tabby-mcp-server by cbusillo - MCP Server

TabbyML MCP Server

MCP server bridge that connects TabbyML to AI assistants for searching and understanding any codebase.

Prerequisites

TabbyML running locally (default: http://localhost:8080)
TabbyML authentication token
Python 3.12+
uv package manager

Setup

# Clone and install
git clone https://github.com/cbusillo/tabby-mcp-server.git
cd tabby-mcp-server
uv sync --all-groups

# Set up your TabbyML token (optional - can also pass via MCP config)
echo "TABBY_TOKEN=your_tabby_token_here" > .env

Quick Setup with Claude CLI

# Add the MCP server to Claude Code
claude mcp add-json tabby-mcp-server '{
  "command": "uv",
  "args": ["run", "--directory", "/path/to/tabby-mcp-server", "tabby-mcp-server"],
  "env": {
    "TABBY_TOKEN": "your_tabby_token_here",
    "TABBY_URL": "http://localhost:8080"
  }
}'

# If using absolute path to uv (recommended)
claude mcp add-json tabby-mcp-server '{
  "command": "/opt/homebrew/bin/uv",
  "args": ["run", "--directory", "/path/to/tabby-mcp-server", "tabby-mcp-server"],
  "env": {
    "TABBY_TOKEN": "your_tabby_token_here"
  }
}'

Environment Variables

TABBY_TOKEN - Required: Your TabbyML authentication token
TABBY_URL - Optional: TabbyML server URL (default: http://localhost:8080)

Tools

list_repositories - List all repositories indexed in TabbyML
search_code - Search for code patterns in indexed repositories
chat_with_context - Chat with TabbyML AI using repository context
search_documentation - Search for information in documentation repositories

Development

uv run test          # Run tests
uv run format        # Format code
uv run lint          # Lint code
uv run update-graphql # Update GraphQL client from TabbyML

Updating GraphQL Client

When TabbyML updates their GraphQL schema or you modify queries:

# Fetch latest schema and regenerate client code
uv run update-graphql

# This command will:
# 1. Fetch the latest schema from TabbyML
# 2. Convert it to GraphQL SDL format
# 3. Generate typed Python client code
# 4. Update src/tabby_mcp_server/graphql_client/

Note: Ensure TabbyML is running and TABBY_TOKEN is set before updating.

Running Integration Tests

Integration tests require a running TabbyML instance:

# Skip integration tests (default)
uv run test

# Run all tests including integration tests
TABBY_TOKEN=your_token pytest tests/test_integration.py -v

# Run specific integration test
TABBY_TOKEN=your_token pytest tests/test_integration.py::TestTabbyIntegration::test_real_tabby_connection -v

Integration tests are marked with @pytest.mark.skipif and will be skipped if:

TabbyML is not running
TABBY_TOKEN is not set
Connection to TabbyML fails

Optional: Git pre-commit hook

Create .git/hooks/pre-commit:

#!/bin/bash
set -e
uv run format
uv run lint
uv run test

Then: chmod +x .git/hooks/pre-commit

Requirements

TabbyML running locally (default: http://localhost:8080)
Python 3.12+ (3.13 compatible)
uv

Documentation Proxy

The tabby-mcp-server includes a built-in documentation proxy to solve TabbyML timeout issues with multi-language documentation sites. See for details.

# Start proxy for any documentation site
uv run tabby-doc-proxy --url https://docs.example.com/

# Configure TabbyML to use: http://localhost:8888/

Example CLAUDE.md for Your Project

Add this to your project's CLAUDE.md to leverage TabbyML for code understanding:

# CLAUDE.md

Project instructions for AI assistance.

## Available Tools

### TabbyML Integration
When working with this codebase, use the TabbyML MCP tools:
- `list_repositories` - View all indexed repositories
- `search_code` - Search for patterns, classes, methods in the codebase
- `chat_with_context` - Ask questions about the code with repository context
- `search_documentation` - Search in documentation repositories

## Code Search Strategy

1. **Before making changes**:
   - Use `search_code` to find similar implementations
   - Use `chat_with_context` to understand complex code sections
   - Use `search_documentation` for API references and guides

2. **Understanding existing code**:
   - List repositories to see what's available
   - Search for usage patterns before implementing new features
   - Ask contextual questions about specific repositories

## Example Workflow

When implementing a new feature:
1. First use `list_repositories` to see available codebases
2. Use `search_code` to find similar implementations
3. Use `chat_with_context` to understand the patterns
4. Check `search_documentation` for best practices
5. Implement following the discovered patterns