mcp-client-for-ollama

A simple yet powerful Python client for interacting with Model Context Protocol (MCP) servers using Ollama, allowing you to harness local LLMs for advanced tool execution.

MCP Client for Ollama (ollmcp)

🎥 Watch this demo as an Asciinema recording

Table of Contents

• Overview
• Features
• Requirements
• Quick Start
• Usage
  • Command-line Arguments
  • Usage Examples
  • How Tool Calls Work
  • ✨NEW Agent Mode
• Interactive Commands
  • ✨NEW Answer Display Modes
  • ✨NEW Input Mode
  • MCP Tools
  • Model Selection
  • Advanced Model Configuration
  • Server Reloading for Development
  • Human-in-the-Loop (HIL) Tool Execution
  • ✨NEW MCP Prompts
  • ✨NEW MCP Resources
  • Performance Metrics
  • ✨NEW History Management
• Autocomplete and Prompt Features
• Configuration Management
• Server Configuration Format
  • Tips: Where to Put MCP Server Configs and a Working Example
• Compatible Models
  • Ollama Cloud Models
• Where Can I Find More MCP Servers?
• Related Projects
• License
• Acknowledgments

Overview

MCP Client for Ollama (ollmcp) is a modern, interactive terminal application (TUI) built for harness engineering, connecting local Ollama LLMs to one or more Model Context Protocol (MCP) servers. By fully supporting the core MCP primitives (tools, prompts, and resources), it provides a controlled terminal space where you steer, and the agent executes. With a rich, user-friendly interface, it lets you safely manage your setup in real time with no coding required. Whether you’re building, testing, or exploring, this client streamlines your workflow with features like fuzzy autocomplete, advanced model configuration, MCP server hot-reloading for rapid development, and strict Human-in-the-Loop safety controls.

Features

• 🤖 Agent Mode: Iterative tool execution when models request multiple tool calls, with a configurable loop limit to prevent infinite loops
• 🌐 Multi-Server Support: Connect to multiple MCP servers simultaneously
• 🚀 Multiple Transport Types: Supports STDIO, SSE, and Streamable HTTP server connections
• 📋 MCP Prompts Support: Browse, invoke, and manage prompts from MCP servers with argument collection, preview, and safe rollback
• 📦 MCP Resources Support: Browse and read contextual data from MCP servers including files, documents, and structured data
• ☁️ Ollama Cloud Support: Works seamlessly with Ollama Cloud models for tool calling, enabling access to powerful cloud-hosted models while using local MCP tools
• 🎨 Rich Terminal Interface: Interactive console UI with modern styling
• 🌊 Streaming Responses: View model outputs in real-time as they’re generated
• 📝 Answer Display Modes: Switch between Plain, Markdown, or Both response views while streaming
• 🛠️ Tool Management: Enable/disable specific tools or entire servers during chat sessions
• 🧑‍💻 Human-in-the-Loop (HIL): Review and approve tool executions before they run for enhanced control and safety
• 🎮 Advanced Model Configuration: Fine-tune 15+ model parameters including context window size, temperature, sampling, repetition control, and more
• 💬 System Prompt Customization: Define and edit the system prompt to control model behavior and persona
• 🧠 Context Window Control: Adjust the context window size (num_ctx) to handle longer conversations and complex tasks
• 🎨 Enhanced Tool Display: Beautiful, structured visualization of tool executions with JSON syntax highlighting
• 🧠 Context Management: Control conversation memory with configurable retention settings
• 🤔 Thinking Mode: Advanced reasoning capabilities with visible thought processes for supported models (e.g., gpt-oss, deepseek-r1, qwen3, etc.)
• 🖼️ Vision Tool Support: Images returned by tools are automatically forwarded to vision-capable models
• 🗣️ Cross-Language Support: Seamlessly work with both Python and JavaScript MCP servers
• 📜 History Management: View full conversation history, export to JSON for backup/analysis, and import previous sessions for continuity
• 🔍 Auto-Discovery: Automatically find and use Claude’s existing MCP server configurations
• 🔁 Dynamic Model Switching: Switch between any installed Ollama model without restarting
• 💾 Configuration Persistence: Save and load tool preferences and model settings between sessions
• 🔄 Server Reloading: Hot-reload MCP servers during development without restarting the client
• ✨ Fuzzy Autocomplete: Interactive, arrow-key command autocomplete with descriptions
• 🏷️ Dynamic Prompt: Shows current model, thinking mode, and enabled tools
• 📊 Performance Metrics: Detailed model performance data after each query, including duration timings and token counts
• 🔌 Plug-and-Play: Works immediately with standard MCP-compliant tool servers
• 🔔 Update Notifications: Automatically detects when a new version is available
• 🖥️ Modern CLI with Typer: Grouped options, shell autocompletion, and improved help output
• ⏹️ Abort Generation: You can abort model generation at any time by pressing ‘a’ during response streaming

Requirements

• Python 3.10+ (Installation guide)
• Ollama running locally (Installation guide)
• UV package manager (Installation guide)

Quick Start

Option 1: Install with pip and run

pip install --upgrade ollmcp
ollmcp

Option 2: One-step install and run

uvx ollmcp

Option 3: Install from source and run using virtual environment

git clone https://github.com/jonigl/mcp-client-for-ollama.git
cd mcp-client-for-ollama
uv venv && source .venv/bin/activate
uv pip install .
uv run -m mcp_client_for_ollama

Usage

Run with default settings:

ollmcp

If you don’t provide any options, the client will use auto-discovery mode to find MCP servers from Claude’s configuration.

Command-line Arguments

ollmcp --install-completion

MCP Server Configuration:

• --mcp-server, -s: Path to one or more MCP server scripts (.py or .js). Can be specified multiple times.
• --mcp-server-url, -u: URL to one or more SSE or Streamable HTTP MCP servers. Can be specified multiple times. See Common MCP endpoint paths for typical endpoints.
• --servers-json, -j: Path to a JSON file with server configurations. See Server Configuration Format for details.
• --auto-discovery, -a: Auto-discover servers from Claude’s default config file (default behavior if no other options provided).

[!TIP] Claude’s configuration file is typically located at: ~/Library/Application Support/Claude/claude_desktop_config.json

Ollama Configuration:

• --model, -m MODEL: Ollama model to use. Default: qwen2.5:7b
• --host, -H HOST: Ollama host URL. Default: http://localhost:11434

General Options:

• --version, -v: Show version and exit
• --help, -h: Show help message and exit
• --install-completion: Install shell autocompletion scripts for the client
• --show-completion: Show available shell completion options

Usage Examples

Simplest way to run the client:

ollmcp

Connect to a single server:

ollmcp --mcp-server /path/to/weather.py --model llama3.2:3b
# Or using short flags:
ollmcp -s /path/to/weather.py -m llama3.2:3b

Connect to multiple servers:

ollmcp --mcp-server /path/to/weather.py --mcp-server /path/to/filesystem.js
# Or using short flags:
ollmcp -s /path/to/weather.py -s /path/to/filesystem.js

Use a JSON configuration file:

ollmcp --servers-json /path/to/servers.json --model llama3.2:1b
# Or using short flags:
ollmcp -j /path/to/servers.json -m llama3.2:1b

Use a custom Ollama host:

ollmcp --host http://localhost:22545 --servers-json /path/to/servers.json --auto-discovery
# Or using short flags:
ollmcp -H http://localhost:22545 -j /path/to/servers.json -a

Connect to SSE or Streamable HTTP servers by URL:

ollmcp --mcp-server-url http://localhost:8000/sse --model qwen2.5:latest
# Or using short flags:
ollmcp -u http://localhost:8000/sse -m qwen2.5:latest

Connect to multiple URL servers:

ollmcp --mcp-server-url http://localhost:8000/sse --mcp-server-url http://localhost:9000/mcp
# Or using short flags:
ollmcp -u http://localhost:8000/sse -u http://localhost:9000/mcp

Mix local scripts and URL servers:

ollmcp --mcp-server /path/to/weather.py --mcp-server-url http://localhost:8000/mcp --model qwen3:1.7b
# Or using short flags:
ollmcp -s /path/to/weather.py -u http://localhost:8000/mcp -m qwen3:1.7b

Use auto-discovery with mixed server types:

ollmcp --mcp-server /path/to/weather.py --mcp-server-url http://localhost:8000/mcp --auto-discovery
# Or using short flags:
ollmcp -s /path/to/weather.py -u http://localhost:8000/mcp -a

Interactive Commands

During chat, use these commands:

Answer Display Modes

The display-mode (dm) command lets you choose how model answers are shown while they stream:

• Plain: Streams the response once as plain text with no final markdown re-render
• Markdown: Renders the response as markdown during streaming with throttled redraws
• Both: Streams plain text first, then renders the completed response again as markdown

Use /display-mode or /dm during chat to open the interactive picker.

Why you might switch modes:

• Plain is the least noisy option if you want minimal redraw or flicker
• Markdown is useful when formatting matters more than raw token-by-token output
• Both gives you fast streaming feedback plus a clean final markdown rendering

Input Mode

The input-mode (im) command controls how you write chat messages:

• Single-line (default): Press Enter to send immediately after typing your message
• Multiline: Press Enter to add a new line, then press Esc followed by Enter to send the entire message when you’re done. This allows for more complex messages with multiple paragraphs or code blocks.
• Ctrl+J also inserts a new line in multiline mode as a reliable fallback across terminals

Use /input-mode or /im during chat to open the interactive picker.

MCP Tools

The tool and server selection interface allows you to enable or disable specific tools:

• Enter numbers separated by commas (e.g. 1,3,5) to toggle specific tools
• Enter ranges of numbers (e.g. 5-8) to toggle multiple consecutive tools
• Enter S + number (e.g. S1) to toggle all tools in a specific server
• a or all - Enable all tools
• n or none - Disable all tools
• d or desc - Show/hide tool descriptions
• j or json - Show detailed tool JSON schemas on enabled tools for debugging purposes
• s or save - Save changes and return to chat
• q or quit - Cancel changes and return to chat

Model Selection

The model selection interface shows all available models in your Ollama installation:

• Enter the number of the model you want to use
• s or save - Save the model selection and return to chat
• q or quit - Cancel the model selection and return to chat

Advanced Model Configuration

The model-config (mc) command opens the advanced model settings interface, allowing you to fine-tune how the model generates responses:

System Prompt

• System Prompt: Set the model’s role and behavior to guide responses.

Key Parameters

• System Prompt: Set the model’s role and behavior to guide responses.
• Context Window (num_ctx): Set how much chat history the model uses. Balance with memory usage and performance.
• Keep Tokens: Prevent important tokens from being dropped
• Max Tokens: Limit response length (0 = auto)
• Seed: Make outputs reproducible (set to -1 for random)
• Temperature: Control randomness (0 = deterministic, higher = creative)
• Top K / Top P / Min P / Typical P: Sampling controls for diversity
• Repeat Last N / Repeat Penalty: Reduce repetition
• Presence/Frequency Penalty: Encourage new topics, reduce repeats
• Stop Sequences: Custom stopping points (up to 8)
• Batch Size (num_batch): Controls internal batching of requests; larger values can increase throughput but use more memory.

Commands

• Enter parameter numbers 1-15 to edit settings
• Enter sp to edit the system prompt
• Use u1, u2, etc. to unset parameters, or uall to reset all
• h/help: Show parameter details and tips
• undo: Revert changes
• s/save: Apply changes
• q/quit: Cancel

Example Configurations

• Factual: temperature: 0.0-0.3, top_p: 0.1-0.5, seed: 42
• Creative: temperature: 1.0+, top_p: 0.95, presence_penalty: 0.2
• Reduce Repeats: repeat_penalty: 1.1-1.3, presence_penalty: 0.2, frequency_penalty: 0.3
• Balanced: temperature: 0.7, top_p: 0.9, typical_p: 0.7
• Reproducible: seed: 42, temperature: 0.0
• Large Context: num_ctx: 8192 or higher for complex conversations requiring more context

Server Reloading for Development

The reload-servers command (rs) is particularly useful during MCP server development. It allows you to reload all connected servers without restarting the entire client application.

Key Benefits:

• 🔄 Hot Reload: Instantly apply changes to your MCP server code
• 🛠️ Development Workflow: Perfect for iterative development and testing
• 📝 Configuration Updates: Automatically picks up changes in server JSON configs or Claude configs
• 🎯 State Preservation: Maintains your tool enabled/disabled preferences across reloads
• ⚡️ Time Saving: No need to restart the client and reconfigure everything

When to Use:

• After modifying your MCP server implementation
• When you’ve updated server configurations in JSON files
• After changing Claude’s MCP configuration
• During debugging to ensure you’re testing the latest server version

Simply type /reload-servers or /rs in the chat interface, and the client will:

1. Disconnect from all current MCP servers
2. Reconnect using the same parameters (server paths, config files, auto-discovery)
3. Restore your previous tool enabled/disabled settings
4. Display the updated server and tool status

This feature dramatically improves the development experience when building and testing MCP servers.

Human-in-the-Loop (HIL) Tool Execution

The Human-in-the-Loop feature provides an additional safety layer by allowing you to review and approve tool executions before they run. This is particularly useful for:

• 🛡️ Safety: Review potentially destructive operations before execution
• 🔍 Learning: Understand what tools the model wants to use and why
• 🎯 Control: Selective execution of only the tools you approve
• 🚫 Prevention: Stop unwanted tool calls from executing
• 🔄 Session Mode: Auto-approve all tools for the current query session
• 🛑 Query Abort: Abort entire query without saving to history

HIL Confirmation Display

When HIL is enabled, you’ll see a confirmation prompt before each tool execution:

Example:

HIL Confirmation Options

When prompted, you can choose from the following options:

• y/yes: Execute this specific tool call
• n/no: Skip this tool call and continue with the query
• s/session: Execute this and all subsequent tool calls for the current query without further prompts
• d/disable: Permanently disable HIL confirmations (can be re-enabled with hil command)
• a/abort: Abort the entire query immediately without saving to history

Human-in-the-Loop (HIL) Configuration

• Default State: HIL confirmations are enabled by default for safety
• Toggle Command: Use /human-in-the-loop or /hil to toggle on/off
• Persistent Settings: HIL preference is saved with your configuration
• Quick Disable: Choose “disable” during any confirmation to turn off permanently
• Session Auto-Approve: Use “session” during confirmation to approve all tools for current query
• Query Abort: Use “abort” during confirmation to immediately stop the query without saving
• Re-enable: Use the hil command anytime to turn confirmations back on

Benefits:

• Enhanced Safety: Prevent accidental or unwanted tool executions
• Awareness: Understand what actions the model is attempting to perform
• Selective Control: Choose which operations to allow on a case-by-case basis
• Flexible Workflow: Session mode for efficient multi-tool queries, individual approval for sensitive operations
• Clean Abort: Stop problematic queries immediately without polluting conversation history
• Peace of Mind: Full visibility and control over automated actions

MCP Prompts

MCP Prompts provide reusable, server-defined conversation starters and context templates. Servers can expose prompts with descriptions, required arguments, and pre-formatted messages that help you quickly start specific types of conversations or inject structured context into your chat.

Features

• 📋 Browse Prompts: View all available prompts from connected servers with descriptions and argument requirements
• ⚡️ Quick Invocation: Use slash syntax to invoke prompts (/server:prompt_name recommended)
• 🔤 Autocomplete: Type / to see prompt suggestions with fuzzy matching
• 📝 Argument Collection: Interactive prompts guide you through required parameters
• 👁️ Preview: Review prompt content before injection to ensure it fits your needs
• 🎯 Flexible Injection: Choose to execute immediately or inject-only (add to history without triggering model)
• 🧠 Context-Aware: Automatically adapts behavior based on whether prompt ends with user or assistant message
• 🔄 Safe Rollback: Automatic history cleanup if you abort or encounter errors
• 💬 Text Content: Supports text-based prompt messages (image/audio/resource support coming soon)

How to Use MCP Prompts

Browse Available Prompts:

/prompts  # or '/pr'

This displays all prompts grouped by server, showing their names, required arguments, and descriptions.

Invoke a Prompt:

/server:prompt_name

For example, if a server named docs provides a “summarize” prompt:

/docs:summarize

If a prompt name is unique across connected servers, you can use the short form:

/summarize

If multiple servers expose the same prompt name, the client will ask you to use the qualified form and suggest valid /server:prompt_name options.

Autocomplete:

• Type / to see all available prompts with descriptions
• Continue typing to filter prompts with fuzzy matching
• Use arrow keys to navigate and press Enter to select

Workflow:

1. Type /server:prompt_name (recommended) or select from autocomplete
2. If the prompt requires arguments, you’ll be prompted to provide them
3. Review the prompt preview showing what will be injected
4. Choose how to use the prompt:
   • y/yes (default): Send the prompt to the model and get a response
     • For prompts ending with a user message: Uses that message as the query
     • For prompts ending with an assistant message: Adds “Please respond based on the above context.” as the query
   • i/inject: Just add the prompt to conversation history without triggering the model (lets you type your own query afterward)
   • n/no: Cancel and return to chat
5. The prompt is injected based on your choice
6. If you abort during model generation (press ‘a’), changes are automatically rolled back

Example:

MCP Resources

MCP Resources provide access to contextual data exposed by MCP servers-files, documents, structured data, and more. Servers can expose resources with metadata (name, description, MIME type) that you can browse and read into your conversation context.

Features

• 📋 Browse Resources: View all available resources from connected servers with URIs, names, MIME types, and descriptions
• 📖 Read Resources: Use @uri syntax to read resource content, standalone or inline within a query
• 📝 Text Content: Full support for text-based resources (markdown, code, logs, etc.)
• 🖼️ Vision Image Support: Image resources (image/*) are automatically forwarded as base64 images to vision-capable models
• 🎯 Context Injection: Resource content is buffered and injected as context alongside your next query
• 🔍 Autocomplete: Type @ to see available resource and template suggestions with fuzzy matching
• 🛡️ Binary Safety: Non-image binary content (audio, video, PDFs, archives) is detected and gracefully skipped with informative messages

How to Use MCP Resources

Browse Available Resources:

/resources  # or '/res'

This displays all resources and templates grouped by server, showing URIs, names, MIME types, and descriptions. Binary resources are marked with a [binary] tag and templates with a [template] tag.

Read a Resource:

@<uri>

For example, to read a file resource:

@file:///path/to/document.md

There are two ways to use @uri:

1. Standalone (buffer then query): Type @uri on its own. The resource is fetched and buffered. Then type your query on the next prompt. The resource content is injected as context automatically.

2. Inline (single turn): Include @uri anywhere inside your query text. The resource is fetched and the query is processed immediately in one step.

Standalone example:

qwen3/show-thinking/6-tools❯ @server://info
✅ Read resource 'get_server_info' (197 chars)

Preview:
This is a simple MCP server with streamable HTTP transport. It supports tools for greeting, adding numbers, generating
random numbers, and calculating BMI. It also provides a BMI calculator prompt.

1 resource(s) buffered. Type your query, or include @another_uri inline.

qwen3/show-thinking/6-tools❯ Next question here

Inline example:

qwen3/show-thinking/6-tools❯ summarize the key features from @server://info
✅ Read resource 'get_server_info' (197 chars)

Preview:
This is a simple MCP server with streamable HTTP transport. It supports tools for greeting, adding numbers, generating
random numbers, and calculating BMI. It also provides a BMI calculator prompt.
[model response]

[!NOTE] 🖼️ Images (image/*) are supported, they are passed directly to vision-capable models as base64 data. Binary Content: The following resource types are not supported as context and will be skipped with an informative message:

• 🎵 Audio - audio/* MIME types
• 📹 Video - video/* MIME types
• 📄 PDFs - application/pdf
• 🗜️ Archives - application/zip, application/octet-stream

Performance Metrics

The Performance Metrics feature displays detailed model performance data after each query in a bordered panel. The metrics show duration timings, token counts, and generation rates directly from Ollama’s response.

Displayed Metrics:

• total duration: Total time spent generating the complete response (seconds)
• load duration: Time spent loading the model (milliseconds)
• prompt eval count: Number of tokens in the input prompt
• prompt eval duration: Time spent evaluating the input prompt (milliseconds)
• eval count: Number of tokens generated in the response
• eval duration: Time spent generating the response tokens (seconds)
• prompt eval rate: Speed of input prompt processing (tokens/second)
• eval rate: Speed of response token generation (tokens/second)

Example:

Performance Metrics Configuration

• Default State: Metrics are disabled by default for cleaner output
• Toggle Command: Use show-metrics or sm to enable/disable metrics display
• Persistent Settings: Metrics preference is saved with your configuration

Benefits:

• Performance Monitoring: Track model efficiency and response times
• Token Tracking: Monitor actual token consumption for analysis
• Benchmarking: Compare performance across different models

History Management

The History Management feature allows you to view, export, and import your conversation history. This is useful for:

• 📜 Full History View: Review all conversations from your current session
• 💾 Export: Save conversations to JSON files for backup or analysis
• 📥 Import: Load previous conversation history to continue where you left off
• 🔄 Portability: Share or transfer conversations between sessions

History Commands

View Full History:

full-history  # or 'fh'

Displays all conversation history from the current session in a formatted view, showing both queries and responses.

Export History:

export-history  # or 'eh'

Exports your current chat history to a JSON file. You can specify a custom filename or use the default timestamp-based name (e.g., ollmcp_chat_history_2026-01-05_143022.json). Files are saved to ~/.config/ollmcp/history/ directory. The command includes file overwrite protection.

Import History:

import-history  # or 'ih'

Imports a previously exported chat history from a JSON file. The command validates the JSON structure to ensure compatibility. Imported history is added to your current conversation context.

History Storage:

• Export location: ~/.config/ollmcp/history/
• Default filename format: ollmcp_chat_history_YYYY-MM-DD_HHMMSS.json
• JSON format includes both queries and responses with proper structure validation

Benefits:

• Session Continuity: Resume conversations across different sessions
• Backup: Keep records of important conversations
• Analysis: Export history for external analysis or review
• Sharing: Share conversation context with team members
• Testing: Import test conversations for development and debugging

Autocomplete and Prompt Features

Typer Shell Autocompletion

• The CLI supports shell autocompletion for all options and arguments via Typer
• To enable, run ollmcp --install-completion and follow the instructions for your shell
• Enjoy tab-completion for all grouped and general options

FZF-style Autocomplete

• Slash-namespace autocomplete for commands and prompts (/)
• Command descriptions shown in the menu
• Case-insensitive matching for convenience
• Centralized command list for consistency
• Plain text query typing is intentionally free of action autocomplete noise

MCP Prompts Autocomplete

• Type / to trigger prompt autocomplete
• Fuzzy matching on prompt names and descriptions
• Supports qualified prompt references like /server:prompt_name
• Shows prompt descriptions in the menu
• Prompt arguments are collected during prompt invocation (not shown in autocomplete rows)
• Terminal-width-aware description truncation

Contextual Prompt

The chat prompt now gives you clear, contextual information at a glance:

• Model: Shows the current Ollama model in use
• Thinking Mode: Indicates if “thinking mode” is active (for supported models)
• Tools: Displays the number of enabled tools

Example prompt:

qwen3/show-thinking/12-tools❯

• qwen3 Model name
• /show-thinking Thinking mode indicator (if enabled, otherwise /thinking or omitted)
• /12-tools Number of tools enabled (or /1-tool for singular)
• ❯ Prompt symbol

This makes it easy to see your current context before entering a query.

Configuration Management

The client supports saving and loading tool configurations between sessions:

• When using save-config, you can provide a name for the configuration or use the default
• Configurations are stored in ~/.config/ollmcp/ directory
• The default configuration is saved as ~/.config/ollmcp/config.json
• Named configurations are saved as ~/.config/ollmcp/{name}.json

The configuration saves:

• Current model selection
• Advanced model parameters (system prompt, temperature, sampling settings, etc.)
• Enabled/disabled status of all tools
• Context retention settings
• Thinking mode settings
• Answer display mode preference
• Tool execution display preferences
• Performance metrics display preferences
• Human-in-the-Loop confirmation settings

Server Configuration Format

The JSON configuration file supports STDIO, SSE, and Streamable HTTP server types (MCP 1.10.1):

{
  "mcpServers": {
    "stdio-server": {
      "command": "command-to-run",
      "args": ["arg1", "arg2", "..."],
      "env": {
        "ENV_VAR1": "value1",
        "ENV_VAR2": "value2"
      },
      "disabled": false
    },
    "sse-server": {
      "type": "sse",
      "url": "http://localhost:8000/sse",
      "headers": {
        "Authorization": "Bearer your-token-here"
      },
      "disabled": true
    },
    "http-server": {
      "type": "streamable_http",
      "url": "http://localhost:8000/mcp",
      "headers": {
        "X-API-Key": "your-api-key-here"
      },
      "disabled": false
    }
  }
}

Tips: where to put MCP server configs and a working example

A common point of confusion is where to store MCP server configuration files and how the TUI’s save/load feature is used. Here’s a short, practical guide that has helped other users:

• The TUI’s save-config / load-config (or sc / lc) commands are intended to save TUI preferences like which tools you enabled, your selected model, thinking mode, display mode, and other client-side settings. They are not required to register MCP server connections with the client.
• For MCP server JSON files (the mcpServers object shown above) we recommend keeping them outside the TUI config directory or in a clear subfolder, for example:

~/.config/ollmcp/mcp-servers/config.json

You can then point ollmcp at that file at startup with -j / --servers-json.

Here a minimal working example let’s say this is your ~/.config/ollmcp/mcp-servers/config.json:

{
  "mcpServers": {
    "github": {
      "type": "streamable_http",
      "url": "https://api.githubcopilot.com/mcp/",
      "headers": {
        "Authorization": "Bearer mytoken"
      }
    }
  }
}

With that file in place you can connect using:

ollmcp -j ~/.config/ollmcp/mcp-servers/config.json

Here you can find a GitHub issue related to this common pitfall: https://github.com/jonigl/mcp-client-for-ollama/issues/112#issuecomment-3446569030

Demo

A short demo (asciicast) that should help anyone reproduce the working setup quickly. This example uses an MCP server example with streamable HTTP protocol usage:

Common MCP endpoint paths

Streamable HTTP MCP servers typically expose the MCP endpoint at /mcp (e.g., https://host/mcp), while SSE servers commonly use /sse (e.g., https://host/sse). Below is an excerpt from the MCP specification (2025-06-18):

The server MUST provide a single HTTP endpoint path (hereafter referred to as the MCP endpoint) that supports both POST and GET methods. For example, this could be a URL like https://example.com/mcp.

You can find more details in the MCP specification version 2025-06-18 - Transports.

Compatible Models

The following Ollama models work well with tool use:

• gemma4
• qwen3.5
• lfm2.5-thinking
• llama3.2
• mistral

For a complete list of Ollama models with tool use capabilities, visit the official Ollama models page.

For models that can also process images returned by tools, see the Ollama vision models page.

Ollama Cloud Models

MCP Client for Ollama now supports Ollama Cloud models, allowing you to use powerful cloud-hosted models with tool calling capabilities while leveraging your local MCP tools. Cloud models can run without a powerful local GPU, making it possible to access larger models that wouldn’t fit on a personal computer.

Supported Ollama Cloud models include for example:

• gpt-oss:20b-cloud
• gpt-oss:120b-cloud
• deepseek-v3.1:671b-cloud
• qwen3-coder:480b-cloud

To use Ollama Cloud models with this client:

1. First, pull the cloud model:

   ollama pull gpt-oss:120b-cloud
   

2. Run the client with your chosen cloud model:

   ollmcp --model gpt-oss:120b-cloud
   

For more information about Ollama Cloud, visit the Ollama Cloud documentation.

How Tool Calls Work

1. The client sends your query to Ollama with a list of available tools
2. If Ollama decides to use a tool, the client:
   • Displays the tool execution with formatted arguments and syntax highlighting
   • Shows a Human-in-the-Loop confirmation prompt (if enabled) allowing you to review and approve the tool call
   • Extracts the tool name and arguments from the model response
   • Calls the appropriate MCP server with these arguments (only if approved or HIL is disabled)
   • Shows the tool response in a structured, easy-to-read format (including image and unsupported-media summaries)
   • If the tool returned images and the current model supports vision, attaches the images to the next LLM message; otherwise displays a warning
   • Sends the tool result back to Ollama
   • If in Agent Mode, repeats the process if the model requests more tool calls
3. Finally, the client:
   • Displays the model’s final response incorporating the tool results

Agent Mode

Some models may request multiple tool calls in a single conversation. The client supports an Agent Mode that allows for iterative tool execution:

• When the model requests a tool call, the client executes it and sends the result back to the model
• This process repeats until the model provides a final answer or reaches the configured loop limit
• You can set the maximum number of iterations using the loop-limit (ll) command
• The default loop limit is 3 to prevent infinite loops

Agent Mode Quick Demo:

Where Can I Find More MCP Servers?

You can explore a collection of MCP servers in the official MCP Servers repository.

This repository contains reference implementations for the Model Context Protocol, community-built servers, and additional resources to enhance your LLM tool capabilities.

Related Projects

• Ollama MCP Bridge - A Python API layer that sits in front of Ollama, automatically adding tools from multiple MCP servers to every chat request. This project provides a transparent proxy solution that pre-loads all MCP servers at startup and seamlessly integrates their tools into the Ollama API.
• MCP Server with Streamable HTTP Example - An example MCP server demonstrating the usage of the streamable HTTP protocol.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• Ollama for the local LLM runtime
• Model Context Protocol for the specification and examples
• Rich for the terminal user interface
• Typer for the modern CLI experience
• Prompt Toolkit for the interactive command line interface
• UV for the lightning-fast Python package manager and virtual environment management
• Asciinema for the demo recording

Made with ❤️ by jonigl