Sourcegraph MCP

MCP server exposing Sourcegraph's AI-enhanced code search capabilities to coding agents.

Contents:

• What this is

  • Features

• Quickstart

  • Prerequisites

  • Configuration

  • Installing and running the server

• Connecting your coding agents

  • Cursor

  • Claude Code

  • Codex CLI

  • Gemini CLI

• MCP tools

  • search

  • search_prompt_guide

  • fetch_content

• Development

  • Linting and formatting

What this is

This MCP server integrates with Sourcegraph, a universal code search platform that enables searching across multiple repositories and codebases. It provides powerful search capabilities with advanced query syntax, making it ideal for AI assistants that need to find and understand code patterns across large codebases.

This is an actively-maintained, detached fork of divar-ir/sourcegraph-mcp.

Features

• Code search: Search across codebases using Sourcegraph's powerful query language

• Advanced query language: Support for regex patterns, file filters, language filters, and boolean operators

• Repository discovery: Find repositories by name and explore their structure

• Content fetching: Browse repository files and directories

• AI integration: Designed for LLM integration with guided search prompts

• Python 3.10+ compatible: Fully tested and working on Python 3.10, 3.11, and 3.12+

Related MCP server: zoekt-mcp

Quickstart

Prerequisites

• A Sourcegraph instance: access to a Sourcegraph instance (i.e., either the sourcegraph.com cloud-hosted remote or a private self-hosted instance)

  Note the Sourcegraph cloud-hosted remote offers a free tier: this is probably the quickest option for first-time/unfamiliar users.

• Python 3.10+

• uv (optional but recommended): offers easier dependency & interpreter management

Configuration

Server configuration is done via environment variables. You can either:

• Add these to a .env file in this repo (using .env.sample as a template)

• Prepend them to the server launch command

Required values

Optional values

Installing and running the server

Method 1: From source using uv (recommended)

1. Clone the repo:

   git clone https://github.com/akbad/sourcegraph-mcp.git
   cd sourcegraph-mcp

2. Install dependencies

   uv sync

3. Run the server

   uv run python -m src.main

Method 2: From source using pip/python

1. Do either of the following:

   • Install via pip directly from GitHub

     pip install git+https://github.com/akbad/sourcegraph-mcp.git

   • Clone the repo source

     git clone https://github.com/akbad/sourcegraph-mcp.git

2. Install and run the server:

   cd sourcegraph-mcp
   pip install -e .
   python -m src.main

Method 3: Using a Docker container

# Pull from GitHub Container Registry
docker pull ghcr.io/akbad/sourcegraph-mcp:latest

# Or build locally
git clone https://github.com/akbad/sourcegraph-mcp.git
cd sourcegraph-mcp
docker build -t sourcegraph-mcp .

# Run the container with default ports...
docker run -p 8000:8000 -p 8080:8080 \
  -e SRC_ENDPOINT=https://sourcegraph.com \
  -e SRC_ACCESS_TOKEN=your-token \
  ghcr.io/akbad/sourcegraph-mcp:latest

# ... or custom ports
docker run -p 9000:9000 -p 9080:9080 \
  -e SRC_ENDPOINT=https://sourcegraph.com \
  -e SRC_ACCESS_TOKEN=your-token \
  -e MCP_SSE_PORT=9000 \
  -e MCP_STREAMABLE_HTTP_PORT=9080 \
  ghcr.io/akbad/sourcegraph-mcp:latest

Connecting your coding agents

Cursor

After running the MCP server, add the following to your .cursor/mcp.json file:

{
  "mcpServers": {
    "sourcegraph": {
      "url": "http://localhost:8080/sourcegraph/mcp/"
    }
  }
}

Claude Code

After running the MCP server, add it to Claude Code using the claude CLI:

claude mcp add --transport http sourcegraph --scope user \
  http://localhost:8080/sourcegraph/mcp/

Verify the server was added:

claude mcp list

Codex CLI

After running the MCP server, add the following to your ~/.codex/config.toml:

[mcp_servers.sourcegraph]
url = "http://localhost:8080/sourcegraph/mcp/"
transport = "http"

Verify the server is configured:

codex mcp list

Gemini CLI

After running the MCP server, add the following to your ~/.gemini/settings.json:

{
  "mcpServers": {
    "sourcegraph": {
      "httpUrl": "http://localhost:8080/sourcegraph/mcp/"
    }
  }
}

Verify the server is configured:

gemini mcp list

MCP tools

This server exposes 3 tools to coding agents:

search

Search across codebases using Sourcegraph's advanced query syntax with support for regex, language filters, and boolean operators.

Parameters:

• query: The search query string (required)

• limit: Maximum number of results to return (optional, default: 30, range: 1-100)

search_prompt_guide

• Generate a context-aware guide for constructing effective search queries based on your specific objective.

• This tool helps AI assistants learn how to use Sourcegraph's query syntax effectively.

Parameters:

• objective: What you're trying to find or accomplish

fetch_content

Retrieve file contents or explore directory structures from repositories.

Parameters:

• repo: Repo path (e.g., "github.com/org/project")

• path: File or directory path within the repo

Development

Linting and formatting

# Check code style
uv run ruff check src/

# Format code
uv run ruff format src/

# Fix auto-fixable issues
uv run ruff check --fix src/