Teleport Docs

You can run a MCP server for Teleport Documentation to perform fast vector searches over the embedded Teleport docs. The MCP server stores embeddings in a Chroma database and provides a tool that executes vector search requests, returning relevant results to your AI client for interpretation. This setup lets you efficiently retrieve documentation content based on semantic similarity without running large language models inside the MCP tool itself.

How to use

Start a MCP server instance and connect your MCP client to it. The server exposes a vector-search capability against the prepopulated database of Teleport docs. You can run the server in a container or locally, depending on your preferred workflow. When you issue a search, you’ll receive the most semantically similar documentation pages from the database, which your AI tool can then interpret and present to users.

How to install

Prerequisites you need installed on your system before running the MCP server:

• Docker (for running the prebuilt MCP container) and/or Python/uv for local development if you prefer building from source.

Set up MCP with a stdio config

Use a standard input/output configuration to run the MCP server locally. The following represents a simple stdio setup using the official container image.

{
  "mcpServers": {
    "teleport-docs": {
      "command": "docker",
      "args": [
        "run",
        "--rm",
        "-i",
        "stevetelelport/teleport-docs-mcp:v0.1.0"
      ]
    }
  }
}

Run a dedicated SSE MCP server (optional)

If you want server-sent events for streaming updates, you can run an SSE-enabled variant. This uses a second container run that exposes port 8282 and starts the server in SSE mode inside the container.

{
  "mcpServers": {
    "teleport-docs_sse": {
      "command": "docker",
      "args": [
        "run",
        "-d",
        "--name",
        "teleport-docs-mcp-sse",
        "-p",
        "8282:8000",
        "stevetelelport/teleport-docs-mcp:v0.1.0",
        "uv",
        "run",
        "main.py",
        "--sse",
        "--host",
        "0.0.0.0"
      ]
    }
  }
}

Local development build and run flow

Build and run the MCP server locally to iterate quickly on changes. The recommended flow is to build the Docker image locally, then run the container in stdio mode for development. If you want to run a containerized SSE instance, use the second configuration shown above.

Rebuild database and dependencies

The vector database is prepopulated. To refresh the data, remove existing indexes and fetch the latest Teleport docs and examples, then regenerate embeddings.

rm -rf docs/pages
rm -rf docs/pages_fixed
cp /path/to/teleport/docs/pages docs/pages
cp /path/to/teleport/examples docs/examples
python3 fix_include.py

rm -rf chroma_index/
python3 embed.py

Available tools

vector_search

Performs a vector search against the Chroma database of Teleport docs and returns the most relevant results to the caller.