eda-agent

eda-agent

MCP server that lets an AI (or any MCP-compatible client) interact with a live Altium Designer session, with KiCad and EasyEDA Pro available as additional backends. It exposes around 600 tools on Altium, covering schematic, PCB, library, project, and design-agent operations, over a persistent DelphiScript bridge. The AI reads the design you currently have open, asks questions about it, and can modify it in place while you watch. The backend is selected at startup, so each user sees only their own tool set.

⚠️ Experimental. Not all tools are extensively tested. Some can crash the Altium DelphiScript engine. See Known limitations before using on any design you haven’t backed up.

Demo

Claude Code with eda-agent designs a round 54 mm, 4-layer RGB LED display board in Altium Designer from a written spec to manufacturing files. It builds a project library from scratch, with a symbol, footprint, 3D model and part numbers for every part. It then draws the schematic, places 61 LEDs in rings, routes the USB pair at 90 ohms, pours stitched ground, gets DRC clean and writes the Gerbers, drill, BOM, pick-and-place and STEP files. A correction round follows: a review prompt lists issues from the first pass, and Claude reroutes the whole USB pair as a coupled differential pair, gives every part a realistic 3D model from its datasheet drawing and finishes the schematics with title blocks. The result would still get a manual tweaking pass before prototyping, but it is already at a very good stage.

Claude Code with eda-agent draws a 12 V to 3.3 V, 2 A step-down supply around the TI TPS62130 in Altium Designer from one sentence. It fetches the datasheet, sizes the inductor, capacitors and feedback divider from it, builds the symbols, draws the schematic and verifies the netlist.

Support the project

Developing eda-agent and testing it against Altium’s scripting engine takes a significant amount of work. If you find eda-agent useful in your work, sponsorship keeps it free and maintained. Questions and bug reports get an answer whether you sponsor or not.

Sponsors

Professional and Company sponsors appear here with their logo, and Backer and Big thanks sponsors by name. Be the first: github.com/sponsors/salitronic.

Dashboard

Two dashboards ship with eda-agent:

• In-Altium status window - a floating Altium-side window showing live status, request count, cumulative Altium-side time, auto-shutdown countdown, and a per-command log with durations. Hide pings filters the 30 s keep-alive traffic; Slow only isolates slow calls. The Config tab (right) sets what the AI may do: edit designs, delete, close documents, and drive menus and dialogs, plus view options. Changes apply live. The Detach button exits the polling loop cleanly; it saves nothing, so unsaved edits stay in the editor.
• Web dashboard - a local browser dashboard at http://127.0.0.1:8766, focused on design review. A Review tab surfaces datasheet / MPN / manufacturer / footprint coverage gauges and an actionable issue queue (missing datasheet, missing MPN, orphan nets, …); Project, Components, Nets, Libraries and Plan tabs give live structured views. Click any component or net to drill into a detail drawer; one click cross-probes it into Altium. Light / dark theme, server-sent-events live feed. It is auto-started by the MCP server - the Open Dashboard button on the in-Altium status window launches the browser.

How it works

• Altium Designer stays open and in full control of your design
• A DelphiScript polling loop runs inside Altium’s scripting engine
• eda-agent (Python, launched by your MCP client) sends commands via file-based IPC
• Altium executes, writes a response, and returns to polling
• You see the changes happen live in Altium

This is not a batch tool that opens a project, runs a script, and exits. It’s a live connection for as long as you want it (conversational design review, guided refactoring, ad-hoc BOM queries, “what nets does this resistor connect to?”), all on the project you currently have open.

Features

• ~600 tools on the default Altium backend (600+ with both registered) across application, project, library, schematic/general, PCB, and design-agent categories
• Generic primitives (obj_query, obj_modify, obj_create, obj_delete, obj_run_process) that work on almost any schematic or PCB object type via late-binding, avoiding per-type handler proliferation
• Every tool takes a list: give nearly any tool items, a list of its own arguments, and the whole list is one call, one request to Altium and one undo step (pcb_place_arc(items=[...]) places twenty arcs that one Undo takes back). The reply gives each item’s result in order. Some tools take their list directly: pcb_place_tracks, pcb_place_vias, pcb_move_components, sch_place_wires, sch_place_components, sch_set_components_parameters, lib_add_pins, proj_get_connectivity_many, sim_attach_primitives. Typical wall-time savings: 10 to 100x on multi-item edits
• Design review snapshot: design_review_snapshot bundles 8 to 12 review reads (project info, components, nets, rules, diff, messages, stats, unrouted, BOM) into a single call. One LLM turn instead of a dozen
• Design-lint sweep: design_lint_report runs 31 audit checks in one IPC pass and returns a structured violation list - schematic-side (component-parameter visibility per class, power-port orientation, floating ports, multi-output / no-driver nets, duplicate designators, off-grid components) and PCB-side (DNP variant components, tented-via ratio, near-miss track endpoints, signal vias without nearby return via, via antennas, removed pad shapes, components outside outline, pads too close to board edge, invalid polygon regions, optional DRC). Each check is also exposed as a standalone audit_* MCP tool; the dashboard’s Status → Health subtab has a one-click Lint panel that calls /api/lint and groups results by Schematic / PCB
• Canonical circuit blocks: design_add_circuit_block folds a whole block into a DesignPlan in one call, allocating refdes, wiring every pin to the right net and tagging power / ground and roles. Twelve of them: decoupling, pullup, pulldown, series_resistor, voltage_divider, rc_lowpass, rc_highpass, led_indicator, crystal, pi_filter, mosfet_low_side, mosfet_high_side. Naming-agnostic: you supply the part identities, it owns only the wiring pattern. design_list_circuit_blocks returns each one’s parameter contract, so the planner never guesses a parameter name
• Datasheet-first discipline: every component-surfacing response (pcb_get_components, proj_get_bom, proj_get_component_info_many, proj_find_component, lib_search, design_review_snapshot, sim_get_readiness) carries a _datasheet_guidance block with per-part vendor search queries. app_attach / app_ping carry a _system_reminder on their first call and every 10th after it (EDA_AGENT_REMINDER_EVERY sets the interval), so every MCP client that connects sees the rule at session start and again later. LLM-fabricated datasheet values are forbidden; WebFetch/WebSearch are called out by name
• Sch <-> PCB netlist crossref: obj_crossref_net(net_name) compares the schematic pin list against the PCB pad list for the same net. Catches ECO drift, stale post-fabrication routing, phantom nets from port/sheet-entry rename conflicts. in_sync flag + sch_only / pcb_only diff
• SPICE simulation workflow: sim_get_readiness audits every component and partitions into ready / needs-primitive / needs-file. sim_attach_primitives sets SpicePrefix + Value on passives. sim_attach_model links a vendor .mdl / .ckt. sim_run dispatches the simulator. Built-in guardrail: never fabricate a SPICE model file, fetch the vendor one
• Focus-independent PCB access: every PCB handler falls back to GetPCBBoardByPath when GetCurrentPCBBoard returns nil (user has a sch tab focused). No more misleading “No PCB document is active” when the PCB is right there
• Fast and compile-cached: persistent polling loop; ~10 ms per call in active mode. SmartCompile caches DM_Compile with a 2 s TTL so a multi-read review pays for one compile instead of a dozen. Explicit proj_force_recompile + proj_get_compile_freshness probes for cases that need a guaranteed-fresh netlist (e.g. after user edits)
• Editor and project settings, read back: board grids and the sheet’s electrical grid, board and schematic editor preferences, auto-save, and project options (net identifier scope, output path, pin swapping). Every value written is read back, and one that did not take is reported in not_applied
• Rigid-flex and route clean-up: split a board into layer stack regions, give each region its own stack (a flex stack with fewer copper layers), add and edit bend lines with radius, angle and fold order, and run Altium’s Gloss or Retrace on chosen nets with the track count and length reported before and after
• Layer stack manager: the whole physical stack with Z heights and finished thickness, dielectric and solder mask settings (Dk, Df, thickness, stiffener), layers inserted above or below a chosen layer, impedance profiles created and set per layer, and each via’s protection structure and back drilling
• Board output and editor settings: the Gerber and print setup saved in a board, Testpoint Style rules, the coarse visible grid factor and each grid’s unit, what pasting keeps and the paste array, confirmations, auto-pan and netlist-load behaviour, display fonts and colours, more schematic preferences, a pad’s drill type, and sheet reference zones for every part, label and port
• Net colour highlighting: give board nets a colour and draw them in it, or copy each net’s project (schematic) colour onto the board so both highlight alike
• Annotation runs silently: proj_annotate designates components without popping the annotate dialog
• Drives the GUI where there is no API: a great deal of Altium exists only as menus and dialogs. app_menu clicks through the real menu bar by name (and lists what a menu holds, so paths are discovered rather than guessed); app_list_open_dialogs reports what is on screen, what it says and whether Altium is stuck; app_ui sets checkboxes, fields, dropdowns and grid rows; app_press_dialog_button presses one button; app_drive_dialogs answers a whole sequence reactively. None of it uses the bridge, only Win32 and the accessible layer, so it keeps working while a modal has the scripting engine blocked, which is exactly when you need it. app_menu with how="drive" fires a menu command and answers its dialogs in the same call, because a command that opens a modal blocks the bridge and a second call would never arrive. The driver decides from what is on screen rather than from a script, stops on any dialog it cannot classify, and gates the press that changes the design behind allow_commit. Dialog text that Altium paints into handle-less controls is recovered by OCR and labelled as read from pixels, not text
• Deferred save for speed: mutations mark documents as modified in memory; disk writes happen on explicit app_save_all, and app_detach saves the documents that were edited. A clean document is never rewritten, and edits are not slowed by a save each
• Two dashboards: an in-Altium floating status window (status, request count, per-command performance, command log, Detach button) and a browser-based web dashboard (127.0.0.1:8766) for design review - datasheet / MPN / footprint coverage gauges, an actionable issue queue, component / net drill-in, one-click cross-probe into Altium, light / dark theme. The whole project view loads in one bundled IPC round-trip (project.dashboard_snapshot); the web dashboard auto-starts with the MCP server
• DelphiScript trap linter: scripts/altium/lint.py (wired into build.py) scans the Pascal sources for known parser hazards - Cardinal() casts, malformed hex literals, empty .Add('') arguments, braces inside comments, fixed-size arrays as function locals, reserved-word identifiers - and fails the build before a bad deploy
• Activity logs: every command is appended to workspace/activity.log (one CSV line per command: timestamp, duration, command name, status, response size and the full response). The bridge also writes bridge_trace.log for IPC-level diagnostics
• Bulk-tool nudge: when a singular tool is hit 2 to 3 times in 10 s, the response carries a _hint_bulk field pointing at the batch variant. Clients that missed the bulk tool in the docstring learn about it at runtime
• Design agent surface: MCP tools (design_get_discipline, design_snapshot_inventory, design_validate_plan, design_audit_schematic, design_validate) that let an MCP-client LLM produce a structured DesignPlan JSON, audit the drawn sheet for layout problems, and validate ERC + connectivity. Datasheet-first, NDA-isolated by construction
• Schematics drawn from a hand template: the LLM plans the sheet the way an engineer draws it, every coordinate on the grid (rails as straight wires, capacitors hanging to ground, a divider whose midpoint runs straight to its pin), writing wires through pin anchors such as U1.EN. sch_preview_template checks the template and draws it to a PNG without touching the sheet; sch_draw_template places the parts, reads their pins back, checks the wiring against those pins and the plan’s nets, refuses a short or a misplaced part before the first wire, then draws the wires, power ports and labels and stamps the part data. On EasyEDA, easyeda_emit_template and easyeda_emit_template_connections do the same in two passes
• Atomic-parts contract: every existing-status Part must carry mpn, footprint, datasheet_url; the inventory snapshot exposes those fields per component; design_validate emits atomic_parts warnings when the contract is missed. Aligns with the KiCad Atomic / Digi-Key Library / atopile / JITX convention
• Schematic audit: design_audit_schematic returns structured {overlaps, wire_crossings, stacked_ports} for the active schematic - pairs of components whose bboxes intersect, wire segments crossing a non-endpoint component body (real Pascal-side Vertex.* + BoundingRectangle.* accessors), and clusters of 3+ rail glyphs of the same net. Each violation carries enough geometry for the planner to compute a corrective move. Programmatic feedback loop without needing a visual snapshot
• Health and doctor preflight: eda-agent health (offline checks: workspace dir, pointer file, bundled scripts) and eda-agent doctor (full preflight talking to Altium: process running, script polling responsive, version match, a read-only command canary, optional --library lib-path checks). --json for machine-readable output
• pip-installable: no admin, no installer, no touching Altium’s config

Requirements

• Python 3.11+
• An EDA tool, one of:
  • Altium Designer (recent versions, AD20+ preferred) - Windows only
  • KiCad 10 with its API enabled (Preferences → Plugins → Enable KiCad API) and the board open in the PCB editor; nothing extra to install
  • EasyEDA Pro with the bundled extension installed (see extensions/easyeda/README.md)

The server picks a backend at startup (EDA_AGENT_BACKEND, default altium), so one install drives any of them. See EDA backends.

Installation

git clone https://github.com/salitronic/eda-agent
cd eda-agent
python -m pip install -e .

or, from PyPI:

python -m pip install eda-agent

A clone gets every commit as it lands and is updated with server_update; a PyPI install gets tagged releases and is updated with python -m pip install --upgrade eda-agent. Each looks for updates where it came from.

Use python -m pip rather than a bare pip. It works even when pip’s Scripts folder is not on your PATH, which is common when Python was installed for you by IT, and it guarantees the package goes into the same Python you will run it with.

Register the server with your MCP client. The binary is eda-agent and runs on stdio; consult your client’s docs for how to add a local stdio-based server. Everywhere this README says eda-agent, python -m eda_agent does exactly the same thing and does not depend on PATH.

Claude Code

claude mcp add altium eda-agent

Adds eda-agent as an MCP server named altium to your Claude Code project config. Use -s user to register it at the user level (available across every project):

claude mcp add -s user altium eda-agent

If eda-agent isn’t on your PATH (“‘eda-agent’ is not recognized”), register it through Python instead. This needs no path at all:

claude mcp add -s user altium -- python -m eda_agent

The -- matters: it tells claude mcp add that everything after it is the command to run, rather than options for claude itself.

To verify the connection: /mcp in a Claude Code session should list altium as connected.

Other MCP clients

The server speaks standard MCP over stdio; any client that accepts a local stdio command will work. Invoke eda-agent (or eda-agent serve) as the subprocess, or python -m eda_agent if it is not on your PATH.

Altium-side scripts

Drop the Altium script project somewhere you can find it:

eda-agent install-scripts

or, if eda-agent is not recognised:

python -m eda_agent install-scripts

Default destination: wherever Altium already loads the script project from (its Global Projects list, for every Altium version installed), or %USERPROFILE%\EDA Agent\scripts\ the first time. Use --dest PATH to put it elsewhere; Altium still talks to the same workspace either way. eda-agent health reports a registration that points at a missing file.

Register the script as a Global Project in Altium (once):

1. Preferences (the gear icon at the top right; DXP → Preferences before AD18) → Scripting System → Global Projects → Install from file
2. Select the Altium_API.PrjScr you just installed

From then on, every Altium startup compiles the script project and the polling loop is one click away:

1. File → Run Script…
2. Expand Altium_API → Dispatcher.pas, select StartMCPServer, click Run

The polling loop starts and your MCP client can drive Altium.

Updating

For as long as it runs, the server looks once an hour, in the background, whether this repository has new commits; the look is a cheap check that the branch has moved, and a full check follows only when it has. When there are new commits, the next tool reply says so and lists them, app_ping and app_context repeat it, and the assistant asks you before installing anything. A check that fails is reported once too. A copy installed with pip from this repository compares the commit it was installed from with the latest one; any other pip install compares its version with the one here. Either says how to upgrade. If you agree, server_update(apply=True) fast-forwards your clone, reinstalls when the dependencies changed, copies the Altium scripts again when they changed and restarts the polling loop. Then reconnect the server (/mcp in Claude Code) to run the new Python. From a terminal, the same thing is:

eda-agent update          # shows what is new, then asks
eda-agent update --check  # only shows what is new

A clone with uncommitted changes, commits of its own or another branch checked out is never touched; the update says why and stops. Set EDA_AGENT_UPDATE_CHECK=0 to stop the background check from reaching the network.

EDA backends

One tool surface, chosen at startup by EDA_AGENT_BACKEND (or --backend).

Backend	Reached through	Status
altium	a persistent DelphiScript bridge	default, most complete
kicad	KiCad’s IPC API and kicad-cli	optional
easyeda	a browser extension you import into EasyEDA Pro	optional

The EasyEDA connection runs the other way round: the editor dials out to this server, so nothing here can start it or make it connect.

Full detail, including what differs between them and what each one cannot do, is in docs/BACKENDS.md.

Finding the right tool

Tools are grouped by the document they act on, and mixing them up is the most common mistake: lib_ acts on a .PcbLib or .SchLib, pcb_ on an open .PcbDoc, sch_ on an open .SchDoc, and obj_ on either, by the object type it is given (eTrackObject acts on the active board). A board tool aimed at a library does not reliably fail. With no board focused it can resolve some other open board and report success for work you never asked for.

Two tools answer different questions about the surface:

• tool_catalog - search by name, category, maturity or interaction. Use it when you know roughly what the operation is called.
• tool_guide - ask in plain words what you are trying to do. It answers the tool and what it needs first, the tool you were probably reaching for and why it acts on a different document, and the short list of things that are genuinely impossible with the reason.

That last answer is the one worth knowing about. Without it there is no way to tell “you missed it” from “it does not exist”, so the same dead ends get investigated repeatedly. An empty result from tool_guide means the guide has nothing on the subject, not that the server cannot do it.

The same split is stated in the server instructions your MCP client receives at startup, so a client that reads them starts out knowing it.

Tool count (clients that cap it)

Some MCP clients limit how many tools a server may expose, or serialize every schema into the model context at startup and slow noticeably. This server registers several hundred. Set EDA_AGENT_TOOLSET=minimal (or pass --toolset minimal) to advertise just two:

• tool_catalog - find an operation by category, maturity, interaction or name, and get its parameters with with_schema=True.
• tool_invoke - run any tool by name with an arguments dict.

Every other tool stays registered and reachable through that pair; only the advertised list shrinks, from several hundred to two.

claude mcp add -s user altium -e EDA_AGENT_TOOLSET=minimal eda-agent

Only families of tools that differ by a mode are advertised as one tool each: the calculators (pcb_calc), the audits (audit), the dimension and 3D body kinds, the menu runners, the dialog drivers and the checkpoints. Each takes a kind and that kind’s arguments. Every member keeps its own name, description and schema, which tool_catalog returns and tool_invoke runs. Everything else keeps a tool of its own; collapsing whole areas into pcb(action=...) style entry points would lose the per-tool descriptions a model finds the right operation by.

The trade-off: in minimal the model no longer sees tool schemas up front, so it must discover before it can act, and an argument mistake surfaces as the target tool’s own error rather than a schema validation message. Call tool_catalog(query=..., with_schema=True) to get a tool’s parameters and required list before invoking it, rather than guessing argument names - some are not what they look like (current_amps, not current_a), and the same tool can differ between backends. Prefer full (the default) unless your client forces otherwise.

Part sourcing

part_search queries every enabled provider and merges the results, each hit attributed to the source that found it; part_fetch then pulls one part’s detail from a provider you name. No provider is enabled by default, and there is no fallback order, so a result always names its source.

The providers, their credentials and their access policies are in docs/PART_SOURCING.md.

Altium 365, company policies, simulators

Altium 365. The altium365_ tools read workspaces, projects and the managed library, and can publish, release and change parameters (only with confirm=True). Two routes: the workspace’s own API, with EDA_AGENT_ALTIUM365_DOMAIN and a token made under Admin > Developer in EDA_AGENT_ALTIUM365_TOKEN; or Nexar, with your Nexar app’s client id in EDA_AGENT_NEXAR_APP_ID and eda-agent altium365-login once (eda-agent altium365-logout forgets it). EDA_AGENT_ALTIUM365_WORKSPACES limits the tools to the workspaces it lists. Nothing Altium 365 returns is written to disk. altium365_check_setup says which route is set up.

Company policies are your files in EDA_AGENT_POLICY_DIR: a naming.json (a template and pattern per footprint, symbol and component) for lib_check_naming and lib_make_name, and rules files in rules/*.json for design_check_company_rules. With none set, nothing is imposed.

Simulators. design_simulate_deck runs a deck in ngspice when it is installed, or in LTspice when you ask for it or it is the only one installed. design_simulators lists what is installed.

Example use cases

Full-project design review

“Do a design review of the PoE front-end. Pull the snapshot, fetch the TPS2372 and TL072 datasheets, and flag anything that doesn’t match.”

One design_review_snapshot call gives the AI project info, design stats, components, nets, rules, diff, messages, board stats, and BOM, plus a datasheet-fetch checklist. The AI then grounds every recommendation in the vendor datasheets it actually pulled. 8 to 12 separate queries collapse into one tool call.

Schematic review

The AI reads your schematic live. Ask it anything a reviewer would:

“List every component connected to the 3V3 rail and flag anything whose datasheet limit is below that.”

“Find all net labels that appear only once across the whole project. Those are probably typos.”

“What’s driving the /RESET net? Walk the connectivity and tell me where it resets and how.”

“Do any two components share a designator prefix with gaps in numbering (e.g. R1, R2, R4)? Re-annotate or tell me what’s missing.”

“Compare the focused schematic to the version from 3 weeks ago. What parameter values changed?”

Behind that, the AI calls tools like obj_query(object_type="eSchComponent", scope="project"), proj_get_connectivity_many(designators=[...]), proj_get_nets(...), obj_modify(...), and so on. You watch Altium repaint as it works.

Sch ↔ PCB drift detection

“Run obj_crossref_net on POE_PG. The PCB seems to have R7 on this net but I’m not sure the schematic still does.”

The response shows sch pins, PCB pads, matched count, and the diff in each direction. A non-empty pcb_only list means the board was fabricated from an earlier schematic revision and a later edit broke the post-ECO merge; catch this before the next ECO push rips routed connections. in_sync: false plus the exact diff tells you which port or sheet-entry rename to undo.

SPICE simulation setup

“Set this schematic up for an AC sweep. Attach SPICE primitives to every passive, fetch vendor SPICE models for the op-amps, and tell me if any part can’t be simulated.”

sim_get_readiness partitions the design into ready / needs_primitive / needs_file. The AI batches primitives onto every passive in one sim_attach_primitives call, searches vendor sites for the IC models, attaches them with sim_attach_model, and reports any holdouts. It will not fabricate a SPICE model file; the rule is baked into the tool response.

Library hygiene

“Open Resistors.SchLib and report every component missing a Value, ManufacturerPart1, or Description parameter. Fill in the missing Description from the datasheet URL if present.”

“Diff our Caps.SchLib against Caps_vendor.SchLib and tell me what’s new or changed.”

“Create a new 48-pin symbol for STM32F411 with this pinout table.”

The last one uses lib_add_pins: one call places the whole pinout in a single transaction instead of 48 LLM turns.

PCB spot-checks

“Any unrouted nets on the board?”

“What’s the total trace length for the USB differential pair, split by layer?”

“Show me all vias on the 12V net and their drill sizes.”

“Run DRC and summarize the violations by severity.”

“What does the Clearance_HV rule actually enforce: clearance value, scope expressions, priority?”

That last one uses pcb_get_rule_properties, which returns the actual numeric gap / widths / impedance targets, not just rule metadata.

Bulk changes

“Every 0402 resistor with value 10k, set its Tolerance parameter to 1% and Voltage to 50V.”

“Rename the net OLD_CS to SPI_CS across every sheet in the project.”

“Move C1-C20 into this 200-mil grid layout pattern.”

A list given to obj_modify, pcb_move_components or sch_place_components (or as items to any other tool) finishes the whole operation in one call and one undo step.

Known limitations

This tool is experimental. Please read this section before using on a design you haven’t backed up.

Bridge changes are checked by Free Pascal and a linter before they ship, which cannot prove Altium’s own DelphiScript engine accepts them: the two differ on which identifiers exist, and an undeclared one faults at runtime rather than at compile time. docs/RELEASE_VERIFICATION.md is the procedure for closing that gap on a release, starting with a self-test that runs inside Altium and needs no document.

UI automation synthesises real keyboard and mouse input

Most of this project talks to Altium through the scripting bridge, which addresses a window handle directly and cannot affect anything else. The app_* UI automation tools are different, and are used where Altium offers no other route: application.execute_menu reports success while invoking nothing, GetMenu returns 0 on Altium’s DevExpress bars, and whole dialogs (Update From Libraries, Preferences, the wizards) have no scripting API at all.

Synthesised input is not addressed to a window. keybd_event and mouse_event are delivered to whatever is active at the instant they fire, and a click lands on whatever is under the pointer. So these tools:

• take focus. Menus and clicks need Altium in front, so running them while you are typing will interrupt you
• cannot be confirmed the way a property write can. A keystroke has no read-back; anything that matters is verified afterwards with a bridge read
• could, without containment, deliver an event to another application if focus or the pointer moved

What contains that: a foreground check runs immediately before every event, including between a key press and its release, and refocuses Altium rather than failing; coordinates are refused unless the point is over a window belonging to Altium’s own process; and no tool accepts a window handle from the caller, so an arbitrary window cannot be addressed. tests/test_foreground_guard.py enforces all three, checking the event guard per line of source rather than per function, because a function that checks once and then emits five events in a loop would pass a naive test while firing four unchecked events.

To switch it off entirely:

EDA_AGENT_UI_AUTOMATION=0

Every synthesised event is then refused. Reading stays available on purpose, because dialog detection is how the rest of the system notices Altium is blocked on a modal.

One case is not solvable: Altium’s menu bar carries entries that are commands rather than menus (Place a Comment, Share, Open Home page, Preferences), and nothing distinguishes them. Measured across all 17 bar items: identical MSAA state including HASPOPUP, identical accDefaultAction, identical UIA control type. Listing such an entry clicks it, and clicking it runs it.

Full detail in docs/ui-automation.md.

Altium DelphiScript engine can crash

Some tool paths trigger DelphiScript compile or runtime errors (“Undeclared identifier…”, “Could not convert variant of type (Dispatch) into type (OleStr)”, etc.). When that happens, the script project halts mid-execution and the polling loop stops responding. You will see one of:

• An Altium error dialog stating the problem
• Your MCP client timing out waiting for a response

Recovery is automatic when the crash happens during a call. The server takes a screenshot of the error box, presses OK, takes a screenshot of the halted line, stops the script and restarts the loop. Other sessions’ calls wait while it does. The call that crashed fails with a message saying the script crashed and where the screenshots and a crash record are (logs/crashes in the workspace). The handler may have applied part of its change, so read the design back before repeating it. For a script found halted some other way, call app_recover_script, or run eda-agent relaunch --restart.

By hand: in Altium Designer, open the script project tab and press the red Stop button in the Script IDE toolbar (equivalently Run > Stop from the menu, or Ctrl+F3; use Ctrl+Pause/Break if the script is stuck in an infinite loop). This stops the halted debugger. Then re-launch the polling loop via File > Run Script… > StartMCPServer > Run.

This is an ongoing reliability effort. Every identified crash is either fixed or guarded. If you hit a new one, the Altium error dialog tells you the exact identifier or line. Opening an issue with that text helps us harden the relevant path.

UNC network paths need current scripts

A UNC path (\\server\team\board.PrjPcb) opens like any other with the current scripts. Scripts deployed before the fix drop one leading backslash, so the file is not found and the error names a path that looks almost right. Run eda-agent install-scripts --force to update them, or use a mapped drive letter (Z:\team\board.PrjPcb) in the meantime.

Text above Latin-1 becomes question marks

Altium’s DelphiScript strings are single-byte, so the bridge carries text as one byte per character. Any character above U+00FF is replaced with ? on the way in, silently. Accented Latin, the micro sign, and the degree sign are all below that boundary and survive; the ohm sign and any CJK text do not, so 10Ω arrives as 10?.

This shows up most often on imported parts: LCSC descriptions are frequently Chinese, and lib_easyeda_import passes the description straight through. If you need those fields readable, set them to a transliteration before importing, or edit them in Altium afterwards.

Altium tool buttons relying on internal scripting pause while the server is running

Altium itself uses DelphiScript internally for many built-in commands (some ribbon buttons, panel actions, menu items). While the eda-agent polling loop is active, those built-in commands may become temporarily unresponsive because Altium’s scripting engine is single-threaded and currently owned by our polling loop.

The polling loop owns the scripting engine for as long as it’s running. While it runs, Altium’s own script-backed buttons sit waiting. The loop exits when either:

• The MCP client calls app_detach (or the dashboard Detach button is clicked); app_detach saves the edited documents (the button saves nothing), the loop exits within ~500 ms, and Altium becomes fully responsive, OR
• 10 minutes of total silence from the MCP client (no commands AND no keep-alive pings) triggers the built-in auto-shutdown

In practice, while an MCP client is attached and sending keep-alive pings every 30 s, the loop will never time out on its own; you need to either have the AI call app_detach or close the MCP client session entirely. After the client disconnects, expect up to ~10 minutes for the loop to auto-exit unless you use Detach to release it immediately.

ECO (sch → PCB update) opens a modal, and there is no silent API

The Altium Schematic API exposes no scripted ECO executor: IECO only records proposed changes, no DM_Execute is documented, and no factory hands a script an IECO. PCB:UpdatePCBFromProject turned out not to be a real process id, so the handler that called it no-opped while reporting success. proj_sync_pcb now invokes WorkspaceManager:Compare, which does the real work and blocks on the change-order dialog.

Two consequences worth knowing before you call it.

It blocks the polling loop until the dialog is answered, so it is not an unattended call on its own. It also refuses while a schematic is focused: Altium answers that case with “Cannot compare a source document against its owner project” and changes nothing. Focus the board with app_set_active_document first.

What it reports is narrower than it looks. components_in_sync counts component presence only, so footprint swaps, designator and parameter edits and net changes are all invisible to it, and dialog_outcome_verified is always false because the handler cannot see which button was pressed.

To drive it end to end, use the GUI tools rather than answering by hand:

app_menu(how="drive", args={"menu_path": "Design|Update PCB Document <board>.PcbDoc", "allow_commit": True})

That clicks the menu and answers the dialogs in one call, which matters because a modal blocks the bridge and a second call would never arrive. It validates before executing, and without allow_commit it presents the change order instead of applying it.

Tools vary in maturity

Not every one of these tools has been exercised on every Altium version or design size. The generic primitives (obj_query, obj_modify, obj_create, obj_delete, obj_run_process) and the core application / project tools are the best-tested. Some PCB modify operations (polygon repour, room creation, align-components) are less battle-tested. Queries are generally safer than mutations.

Timeout and server lifecycle

The server has three independent timeout mechanisms:

1. Per-command timeout (Python side)

When the MCP client calls a tool, the Python bridge writes a request file and waits up to 10 seconds by default for a response. Fast queries typically complete in under 100 ms, so a 10 s ceiling surfaces stalls quickly while leaving plenty of margin for real work. Long-running tools that are expected to take longer (app_save_all, pcb_get_unrouted_nets) set their own larger timeouts internally.

Each request is published to its own request_<id>.json file; Altium replies in response_<id>.json with the matching ID. The bridge’s keep-alive thread and MCP-client calls each use their own request IDs, so responses never race. The older single-response.json channel was retired in IPC v2.

2. Server auto-shutdown (Altium side)

The DelphiScript polling loop auto-stops after 10 minutes of inactivity (AutoShutdownMs = 600000). If the MCP client disconnects and the keep-alive pings stop arriving, the server releases Altium’s scripting engine after ten minutes and StartMCPServer returns. To resume, re-launch via File → Run Script… → StartMCPServer → Run.

3. Python keep-alive pings

While an MCP client is attached, the Python bridge pings Altium every 30 seconds so the 10-minute auto-shutdown never fires mid-session. The sequence:

• AI issues command A → Altium busy, then idle
• 30 s later, Python pings → Altium responds “pong”, idle timer resets
• 10 min later, still no AI activity and no ping → Altium auto-shuts down

In practice: the server stays alive as long as an MCP client is connected, and exits cleanly ~10 minutes after the client fully disconnects. No manual stop needed in the common case. For a hard exit, the AI calls app_detach, which saves the documents that were edited, leaves clean ones untouched, and returns control to Altium within ~500 ms. The Detach button on the dashboard window stops the loop without saving.

Several Claude sessions on one Altium

Every session runs its own eda-agent server, and all of them talk to the one script loop in Altium. Request ids carry a session tag and a sequence number. A session working alone is served straight through, in the order it sent its requests, with no time slots. When more than one session has requests waiting, the session being served keeps going for up to 30 seconds (SESSION_SLOT_MS), then the next waiting session gets one slot, and so on in turn, so no session gets a second slot before every waiting session has had one. A command already running is never interrupted: slots change between commands, so one long command (a large repour, say) still holds everyone else up until it finishes. While sessions contend, Altium refreshes queue.json in the workspace, and a waiting client keeps waiting instead of reporting the loop as dead.

Each session also keeps its own document. Most tools act on the focused document, and any session can move the focus between two calls of another. So every server sends, with each command, the document its session last had in front, and Altium puts that document back in front before running the command, in the same step so nothing can come between. A command’s reply carries session_document when that happened. This holds even when you click to another document yourself: the next command of a busy session brings its own back, because a write on a document the session never chose is worse than a view that moves. To point a session at the document you are looking at, tell it to; app_get_active_document reports what is really in front, and focusing a document makes it the session’s.

Things that affect every session at once: a deploy, a relaunch or a restart of Altium, and app_detach, which saves every document a tool edited, including another session’s. To restart the loop for a deploy, run eda-agent install-scripts --force and then eda-agent relaunch --restart. It waits for the command in progress to finish, and other sessions’ calls wait through the restart instead of failing, so nobody needs to pause. A session whose eda-agent server started before the deploy gets server_outdated on its replies until it reconnects (/mcp in Claude Code). To stop the loop without saving, put an empty file named stop in the workspace.

Why this matters for Altium UI responsiveness

The polling loop goes into idle mode after a second or two of no MCP commands (150 empty polls). In idle mode it polls every 30 ms with a ProcessMessages yield in between, so Altium’s UI stays responsive continuously. In active mode the loop polls every 10 ms (ProcessMessages every 5th tick), giving sub-50 ms round-trip latency for back-to-back commands. For a full release, call app_detach or click Detach on the dashboard.

Tool reference

docs/TOOL_REFERENCE.md lists every tool with its arguments, its maturity (offline / simulator / live-verified) and its interaction badge, flagging the ones that open a blocking dialog or leave work incomplete. It is generated by python scripts/gen_tool_reference.py, so it cannot drift from the code.

At runtime tool_catalog serves the same data, filtered by category, maturity, interaction or substring, and tool_invoke calls anything it lists. That pair is the whole advertised surface under EDA_AGENT_TOOLSET=minimal.

Architecture

    +-----------------------------+
    |    MCP-compatible client    |
    +-----------------------------+
            |              ^
            v              |
       tool call       tool result
       (JSON-RPC)      (JSON-RPC)
            |              |
            v              |
    +-----------------------------+
    |     eda-agent (Python)      |
    | application / project / lib |
    | / generic / pcb / design    |
    |              |              |
    |     Altium bridge (IPC)     |
    +-----------------------------+
            |              ^
            v              |
   request_<id>.json   response_<id>.json
            |              |
            v              |
    +-----------------------------+
    |      Altium Designer        |
    |  DelphiScript polling loop  |
    |     (Altium_API.PrjScr)     |
    +-----------------------------+

All intelligence lives in Python. The DelphiScript side is a pass-through layer for object iteration, property access, and process execution.

CLI

Command	Purpose
eda-agent	Start the MCP server on stdio (what the MCP client calls)
eda-agent serve	Explicit form of the above
eda-agent --no-dashboard / eda-agent --headless	MCP server only, no web dashboard. Required by strict-stdio MCP clients (Codex, etc) that can’t tolerate the dashboard thread. Also via env var: EDA_AGENT_DISABLE_DASHBOARD=1 or EDA_AGENT_HEADLESS=1.
eda-agent scripts-path	Print path to bundled DelphiScript sources
eda-agent install-scripts [--dest PATH] [--force]	Copy scripts to every folder Altium loads them from, or to PATH
eda-agent relaunch [--restart] [--force] [--dry-run]	Have the running Altium start the polling loop from the installed script project; --restart stops the running loop after its current command and starts it again (after install-scripts), and other sessions’ calls wait through it
eda-agent update [--check] [--yes] [--no-relaunch]	Show the new commits on the repository and, once you agree, update this clone, the Altium scripts and the polling loop
eda-agent easyeda-extension path|build [--out DIR]	Print where the EasyEDA extension source is, or build the .eext to import into EasyEDA Pro (needs Node.js)
eda-agent dashboard [--host H] [--port P]	Start the local web dashboard on its own (http://127.0.0.1:8766)
eda-agent stop-dashboard	Stop a dashboard that was launched detached
eda-agent calc [NAME] [KEY=VALUE...] [--json]	Run an engineering calculator; with no name, list them
eda-agent review --offline <file> [--json/--sarif] [--fail-on ...]	Offline component-level design review of a .SchDoc/.PrjPcb (no Altium). Opt-in (--offline or EDA_AGENT_HEADLESS_REVIEW=1); exit 1 on findings at/above --fail-on - a hardware-CI gate
eda-agent bom --offline <file> [--csv/--json]	Offline consolidated BOM from a .SchDoc/.PrjPcb (no Altium). Opt-in
eda-agent netlist --offline <file> [--json] [--fail-on ...]	Offline geometric netlist reconstruction + connectivity ERC (single_pin_net, net_short) from a .SchDoc (no Altium). Opt-in; exit 1 on findings at/above --fail-on
eda-agent altium365-login	Sign in to Nexar in the browser for the altium365_ tools (needs EDA_AGENT_NEXAR_APP_ID)
eda-agent altium365-logout	Forget the saved Nexar login
eda-agent sdk-reach [--detail N] [--json]	How many of the properties Altium’s SDK declares on each object type obj_query and obj_modify reach (needs the SDK index in reference/altium_sdk)
eda-agent health	Fast offline preconditions: workspace dir + writable, pointer file + matches config, bundled scripts findable, bridge constructable. Exit 0 = clean, 1 = critical fail
eda-agent doctor [--library PATH]... [--json]	Full preflight talking to Altium: all health checks plus process running, script polling responsive, script-version matches bundled, a read-only command round-trip, optional --library lib reachability checks (no hardcoded paths; repeat the flag for each lib you want tested)

Configuration

Workspace (used for IPC files between Python and Altium):

• Default: %USERPROFILE%\EDA Agent\workspace\
• Override: set EDA_AGENT_WORKSPACE environment variable
• The DelphiScript side reads the resolved path from C:\ProgramData\eda-agent\workspace-path.txt, which Python writes at startup and on every install-scripts run

UI automation (synthesised keyboard and mouse input, used for the parts of Altium with no scripting API):

• Default: on
• Disable: set EDA_AGENT_UI_AUTOMATION=0 (also false, no, off). Anything else leaves it on, so a typo cannot silently disable it
• Read at call time, so it takes effect on the next event rather than the next restart
• See UI automation synthesises real keyboard and mouse input for what it is for and what it can do

Coordinates throughout the API are in mils (1 mil = 0.0254 mm).

Development

pip install -e ".[dev]"
python -m pytest tests/ -q

The test suite includes a Python Altium simulator for end-to-end integration tests, Free Pascal cross-validation that runs the actual DelphiScript functions against Python mirrors, and a regression suite for previously encountered edge cases.

Rebuild the monolithic DelphiScript file after editing sources under scripts/altium/:

cd scripts/altium
python build.py

Project layout

eda-agent/
├── src/eda_agent/          Python package
│   ├── bridge/             Communication with each editor (Altium IPC, EasyEDA WebSocket)
│   ├── schemas/            Pydantic IPC envelope + per-command schemas
│   ├── tools/              MCP tool implementations
│   ├── core/               EDA-agnostic design snapshot and review
│   ├── design/             Design agent: plan / inventory / discipline / executor / validator
│   ├── layout/, placement/, route/   Board geometry, rules, placement and routing engines
│   ├── libimport/          Part providers and vendor library import
│   ├── render/, export/    Drawing designs to SVG/PNG; file-format exporters
│   ├── fileio/, sim/       Offline readers (opt-in fallback); simulation results and settings
│   ├── ui/                 Windows UI automation for dialogs and menus
│   ├── web/                Local web dashboard
│   ├── diag/               Health and doctor checks
│   ├── cli.py              CLI subcommands
│   └── server.py           MCP server entry point
├── scripts/altium/         DelphiScript sources (dev source of truth)
│   ├── Main.pas, Utils.pas, Dispatcher.pas, …
│   └── Altium_API.PrjScr   Altium script project
├── extensions/easyeda/     The EasyEDA Pro extension (the editor half of that backend)
├── skills/                 Client skill files (autodesign)
├── docs/                   Backends, tool reference, release verification and other guides
└── tests/                  Python + Free Pascal test suite

At wheel build time scripts/altium/ is copied into src/eda_agent/scripts/ inside the wheel (via Hatchling force-include), so eda-agent install-scripts always finds the scripts.

Troubleshooting

“Altium Designer is not running”: open Altium before invoking MCP tools.

“Script not responding” / MCP tools time out: confirm the script project is loaded and StartMCPServer is running. Re-launch it via File > Run Script… > StartMCPServer > Run. Check %USERPROFILE%\EDA Agent\workspace\ is writable.

Altium error dialog “Undeclared identifier: …” or “Could not convert variant…”: a DelphiScript crash in one of the bridge handlers. In Altium’s Script IDE toolbar, press the red Stop button (or Run > Stop / Ctrl+F3; use Ctrl+Pause/Break if the script is stuck in an infinite loop) to halt the debugger. Then re-launch the polling loop via File > Run Script… > StartMCPServer > Run. If the identifier is one the scripts themselves define, Altium is running an older compile of a unit: run app_ping, which names it in stale_units, then close the script project in the Projects panel (right-click Altium_API.PrjScr, Close Project), reopen it and run StartMCPServer. Otherwise report the identifier or error text as an issue.

Some Altium buttons don’t respond while the server is running: expected while the AI is actively issuing commands. Built-in Altium functions that depend on DelphiScript wait for the polling loop to yield. The loop enters an idle/yield mode a second or two after the last AI command; if a button is still unresponsive after that, call app_detach from the MCP client to fully release the scripting engine.

Altium’s main window disappears while the status window stays up: the main window has been found hidden, not minimized, with no taskbar button. The bridge checks after every reply, keep-alive pings included, and shows it again without taking the focus, so it comes back within about 30 s. Each time, bridge_trace.log gets a FRAME_SHOWN line naming the last command, and UI tools log UI tool=<name> lines so they can be named too. If it keeps happening, report that line as an issue.

Command timeouts on very large boards: default is 10 s so stalls surface fast. Tools known to take longer (app_save_all, pcb_get_unrouted_nets) set their own, longer timeouts. If you hit a timeout on a custom long-running operation, embed the bridge directly and pass a higher timeout= to send_command_async. The polling loop itself adapts (10 ms active, 30 ms idle) so it doesn’t add latency.

License

Apache License 2.0. See LICENSE and NOTICE.

Disclaimer

Use at your own risk. eda-agent drives Altium Designer programmatically and can modify, save, or delete design data. An AI client operating it can issue rapid, irreversible changes. Before using this tool on any design:

• Back up your project. Commit to version control, copy the folder elsewhere, or both. Do not rely solely on Altium’s own history.
• Expect the possibility of data loss, corrupted documents, or Altium crashes, especially on large boards, unusual object configurations, or untested API paths.
• Review automated changes before saving. Prefer working on a branch or a copy until you have trust in a given workflow.

This software is provided “as is”, without warranty of any kind, express or implied. The authors and contributors are not liable for any damage to your designs, projects, data, or installation.

This project is not affiliated with, endorsed by, or sponsored by Altium Limited, the KiCad project, or EasyEDA. “Altium” and “Altium Designer” are trademarks of Altium Limited; “KiCad” and “EasyEDA” are trademarks of their respective owners. eda-agent is an independent community tool that interoperates with each of these applications through its own published API: Altium Designer via its scripting API, KiCad via its IPC API and command line, and EasyEDA Pro via its extension API.

About

eda-agent was created and is maintained by George Saliba, an independent electronics design engineer. Thanks to everyone who has contributed, see the contributors.

If you need hardware designed, I take on design work, see salitronic.com.