docs-mcp-server is a multi-tenant MCP (Model Context Protocol) server focused on indexing and serving documentation search results. It lets you ingest documentation from websites, Git repositories, and local Markdown folders, then exposes tools to AI clients for fast, context-rich retrieval. Core MCP tools exposed include list_tenants, find_tenant, describe_tenant, root_search, and root_fetch, which let clients discover tenants, inspect their configuration, perform search queries against the tenant’s indexed docs, and fetch full documents with source URLs. The server is designed to support multiple tenants with scheduler-driven sync so that sources stay fresh. To start using the server locally, run the module that hosts the MCP app and point your MCP client at the host/port you configure; for example, a local run might use port 8000 and host 0.0.0.0. Once running, refer to docs_reference/mcp-tools.md for the exact MCP tool contracts and example payloads.

Typical workflow:

• Start the server using the provided module entry point (see installation instructions).
• Use the MCP tools to manage tenants (list_tenants, find_tenant, describe_tenant) and to perform root-level searches (root_search) and fetches (root_fetch).
• Integrate with an MCP client (Copilot, Claude-compatible client, or a custom tool) by pointing it at the server endpoints and using the MCP contracts described in the project docs.

Prerequisites:

• Python 3.12 or newer
• Git
• Optional: Virtual environment tooling (venv, virtualenv) for isolation

Install and run locally:

1. Clone the repository git clone https://github.com/pankaj28843/docs-mcp-server.git cd docs-mcp-server

2. Create and activate a virtual environment (recommended)

   macOS/Linux

   python -m venv venv source venv/bin/activate

   Windows

   venv\Scripts\activate.bat

3. Install dependencies pip install -r requirements.txt

4. Prepare deployment (optional guidance)

   • If you have deployment.json or deployment.json.example, copy it and adjust tenant sources, sync schedules, and MCP endpoints as needed.

5. Run the server

   • Start the MCP server module (adjust host/port as needed) python -m docs_mcp_server.app

6. Verify the server is reachable

   • Open http://localhost:8000/mcp (or your configured host/port) and test MCP tool endpoints using an MCP client.

Notes:

• If you are using a custom deployment configuration, ensure the deployment.json (or equivalent) is accessible by the server and correctly references tenants and sources.
• You can also run internal maintenance scripts for syncing tenants, as described in the repository documentation.

Tips and common issues:

• Ensure Python 3.12+ is used to match the project’s dependencies.
• If you encounter import errors, make sure your virtual environment is active and that dependencies from requirements.txt are installed.
• When running multiple tenants, monitor resource usage as indexing can be memory-intensive depending on source volume.
• Use the MCP tools documentation (docs/reference/mcp-tools.md) as the canonical contract reference for payload shapes and responses.
• Environment variables can customize host/port, deployment paths, and endpoints; keep them in a deployment.json or the equivalent environment configuration for your runtime.
• If you plan to run behind a reverse proxy, configure appropriate TLS termination and ensure the MCP endpoint path matches your client expectations (e.g., /mcp).