Godot Docs MCP

demo.png

This project is [hosted on Cloudflare using their Agents framework](https://developers.cloudflare.com/agents/)

[!WARNING] Due to the popularity of this project, a rate limit of 30 requests per 60 second has been added. If you want to avoid that, create a free Cloudflare account yourself and deploy this project.

Look up documentation in Godot using fuzz search. Supports stable, latest, 4.6, 4.5, 4.4, and 4.3 versions. The default version is "stable".

Tools

search_docs (searchTerm: string, version: "stable" | "latest" | "4.6" | "4.5" | "4.4" | "4.3" = "stable")

Search the Godot docs by term. Will return URLs to the documentation for each matching term. The resulting URLs will need to have their page content fetched to see the documentation.

get_docs_page_for_term (searchTerm: string, version: "stable" | "latest" | "4.6" | "4.5" | "4.4" | "4.3" = "stable")

Get the Godot docs content by term. Will return the full documentation page for the first matching result.

Configure the MCP server

To use the hosted HTTP server:

{
  "mcpServers": {
    "godot": {
      "type": "http",
      "url": "https://godot-docs-mcp.j2d.workers.dev/mcp"
    }
  }
}

Or, to connect to the MCP server using a stdio server:

{
  "mcpServers": {
    "godot": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://godot-docs-mcp.j2d.workers.dev/mcp"
      ]
    }
  }
}

How this works

The docs site uses a frontend search tool to handle the docs search. There is a file called searchindex.js in the docs site that contains an index of all the pages (URLs and titles, not content) on the site.

This project takes advantage of that in the following ways:

• downloads each of those searchindex.js files for each version of the docs
• converts the searchindex.js to a searchindex.js.json that is just json we need
• indexes that new json using lucaong/minisearch
• when a docs page is requested, the URL for the page is converted from HTML to markdown

Local development

MCP server

npm install
npm run dev

Then, set up your tool:

{
  "mcpServers": {
    "godot": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "http://localhost:8787/mcp"
      ]
    }
  }
}

To debug the server, you can use this browser tool:

# in another tab
npx @modelcontextprotocol/inspector

Then open http://localhost:6274/#tools.

You can also use https://www.mcpplayground.io/ to look at the tools in the live HTTP server.

Generating the docs

# create the folders
echo "stable,latest,4.6,4.5,4.4,4.3" | string split "," | xargs -I {} mkdir -p "src/indexes/{}"
# download the files
echo "stable,latest,4.6,4.5,4.4,4.3" | string split "," | xargs -I {} curl -o "src/indexes/{}/searchindex.js" "https://docs.godotengine.org/en/{}/searchindex.js"
# convert the .js to a .json
lsd src/indexes/*/searchindex.js | xargs -n1 fish -c 'cat "$argv" | sd "Search.setIndex\(" "" | sed \'$ s/.$//\' | jq \'.docnames | to_entries | map({id: .key, name: .value, category: .value | split("/") | first, url: "/\\(.value).html"})\' > "$argv.json"'

Deploy to Cloudflare

To deploy your own instance (recommended to avoid rate limits):

1. Create a free Cloudflare account
2. Install and authenticate Wrangler:

npm install -g wrangler
wrangler login

1. Change the ratelimits settings in the wrangler.jsonc.
2. Clone and deploy:

git clone https://github.com/your-repo/godot-docs-mcp.git
cd godot-docs-mcp
npm install
npm run deploy

1. Update your server URL:

After deployment, update your MCP config with your worker URL:

{
  "mcpServers": {
    "godot": {
      "type": "http",
      "url": "https://godot-docs-mcp.YOUR-SUBDOMAIN.workers.dev/mcp"
    }
  }
}

Adjusting the Rate Limit

The default rate limit is 30 requests per 60 seconds. To increase or disable it, edit wrangler.jsonc:

Increase the limit:

"ratelimits": [
  {
    "name": "MCP_RATE_LIMITER",
    "namespace_id": "1001",
    "simple": {
      "limit": 1000,
      "period": 60
    }
  }
]

Disable rate limiting entirely: Remove the entire ratelimits array from wrangler.jsonc.