krpc-mcp

An MCP server that gives AI models access to the full kRPC Python documentation.

What is this?

If you use AI assistants (Claude, ChatGPT, Codex, etc.) to write kRPC Python code for Kerbal Space Program, the models often guess or hallucinate API calls because kRPC's documentation is uncommon in their training data. This MCP server solves that by letting the AI query the real documentation before writing code.

The server crawls and indexes all kRPC Python API pages, caches them locally, and exposes them through 4 search/retrieval tools via the Model Context Protocol.

Quick Start

Pick your AI client below and run the setup command. That's it -- the AI will automatically have access to kRPC docs.

Claude Code (CLI)

claude mcp add krpc-mcp -- uvx krpc-mcp

OpenAI Codex CLI

codex mcp add krpc-mcp -- uvx krpc-mcp

Or add it manually to ~/.codex/config.toml:

[mcp_servers.krpc-mcp]
command = "uvx"
args = ["krpc-mcp"]

Claude Desktop

Add this to your claude_desktop_config.json:

OS	Config file location
macOS	~/Library/Application Support/Claude/claude_desktop_config.json
Windows	%APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "krpc-mcp": {
      "command": "uvx",
      "args": ["krpc-mcp"]
    }
  }
}

Cursor

Add this to .cursor/mcp.json in your project root:

{
  "mcpServers": {
    "krpc-mcp": {
      "command": "uvx",
      "args": ["krpc-mcp"]
    }
  }
}

Other MCP clients

Any MCP client that supports stdio transport can use this server. The command is:

uvx krpc-mcp

Alternative Installation

If you don't have or don't want to use uvx, there are other ways to install.

pip install from GitHub

pip install git+https://github.com/Ljove02/krpc-mcp.git

Then use krpc-mcp as the command in any of the configs above (instead of uvx krpc-mcp).

Manual setup (for development)

git clone https://github.com/Ljove02/krpc-mcp.git
cd krpc-mcp
python -m venv .venv

Activate the virtual environment:

# macOS / Linux
source .venv/bin/activate

# Windows (PowerShell)
.\.venv\Scripts\Activate.ps1

# Windows (cmd)
.venv\Scripts\activate.bat

Install and run:

pip install -e .
krpc-mcp

Available Tools

Once connected, your AI assistant gets these tools:

Tool	Parameters	Description
search_docs	query (str), limit (int, default 5)	Search all indexed kRPC Python docs by keyword. Returns matching pages with snippets.
get_doc_page	slug_or_url (str)	Get the full content of a specific docs page. Accepts a slug like python/api/space-center/vessel.html or a full URL.
get_api_member	service (str), class_name (str), member (str)	Look up a specific API member. For example: service="SpaceCenter", class_name="Vessel", member="orbit". Uses fuzzy matching.
reindex_docs	force (bool, default false)	Manually trigger a re-crawl of the kRPC documentation.

Example: Real Usage

Here's what it looks like when an AI assistant uses krpc-mcp. In this test, Claude Code was asked about vessel launch telemetry data.

User prompt

"Give me info about all kinds of data we can get from a launching vessel -- speed, altitude, g-force. Give me 5 examples in detail so I can cross-check with the docs."

What the AI did behind the scenes

The AI automatically called the MCP tools to look up real documentation before answering:

Tool call:  search_docs(query="vessel flight telemetry speed altitude", limit=5)
Result:     3 matching pages found, including Flight and Vessel API references

Tool call:  search_docs(query="g force flight data", limit=5)
Result:     Matched the Flight API page

Tool call:  get_doc_page(slug_or_url="python/api/space-center/flight.html")
Result:     Full Flight class documentation (104 lines of API details)

Tool call:  get_doc_page(slug_or_url="python/api/space-center/vessel.html")
Result:     Full Vessel class documentation (173 lines of API details)

AI response (sourced from real docs)

#	Property	Class	Returns	Description
1	Flight.speed	Flight	float	Speed in m/s in the given reference frame. Pass an appropriate reference frame (e.g. vessel.orbit.body.reference_frame) to get meaningful values.
2	Flight.mean_altitude	Flight	float	Altitude above sea level in meters. Related: surface_altitude, bedrock_altitude, elevation.
3	Flight.g_force	Flight	float	Current G-force in multiples of 9.81 m/s². Useful for monitoring crew safety and structural stress during launch.
4	Flight.dynamic_pressure	Flight	float	Dynamic pressure (Q) in Pascals. Equal to 0.5 * air_density * velocity². Critical for identifying Max Q during ascent.
5	Flight.mach	Flight	float	Speed as a multiple of the speed of sound. Related: speed_of_sound, true_air_speed, terminal_velocity.

All properties are read-only and available in the Flight game scene. The AI pulled every detail -- types, descriptions, related members -- directly from the indexed docs instead of guessing.

How It Works

1. On first use, the server crawls all Python documentation pages from https://krpc.github.io/krpc/python.html
2. It extracts page content and API member signatures/descriptions
3. Everything is cached locally at ~/.cache/krpc-mcp/
4. The cache auto-refreshes every 24 hours
5. You can force a refresh anytime with the reindex_docs tool

License

MIT