rune — Claude Code MCP server: what it does, how to install
rune — an MCP server that scans your project into a graph of facts, each tied to a file and line. Answers code questions with ranked findings and citations.
MIT ★ 10 updated 2026-09-30
Open on GitHubWhat rune does
Rune parses project files into a graph: imports, function calls, Express and Next.js routes, React and Vue components. Every fact points to a file, line, and snippet. When you ask a question, the server retrieves relevant facts, ranks them, and sends them to your model — not its own.
Your model writes an explanation, and Rune verifies every statement: if it cites a fact the server did not retrieve, or quotes text absent from the cited evidence, that statement is dropped. Without a configured model (Anthropic, OpenAI, OpenAI-compatible API, or local via Ollama), the server prints facts and setup instructions.
Understands JavaScript, TypeScript, shell, Lua, config formats (TOML, YAML, JSON), and markdown. Other languages yield sparser results. Detects hardcoded secrets, risky shell execution, GitHub Actions issues, and dependency problems. The graph lives in .rune/ inside your project — add it to .gitignore.
What it helps with
- understand how routing or architecture works in a project
- trace where a function or component is imported from
- find hardcoded secrets and dangerous shell commands in code
- get file:line citations for a model's answer about your codebase
Example prompts for the agent
- How does routing work in this project?
- Find all calls to the
authenticatefunction. - Give me five architectural insights for the current project.
How to install rune
Install
The MCP registry has no launch command for it: see the repository README
Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about rune
How to install rune in Claude Code?
On Linux (x64) and macOS (Apple Silicon): curl -fsSL https://raw.githubusercontent.com/thecolourfoundation/rune/main/install.sh | sh. On Windows, download rune-windows-x64.exe from the latest release and run it from a terminal. To connect as an MCP server, add to your client config: {"command": "rune", "args": ["serve", "/path/to/project"]}.
What does rune do?
Scans a project into a graph where every fact is tied to a file and line. Answers questions about code with a selection of relevant snippets and citations that your model can explain. Verifies that each model statement is backed by retrieved facts.
Which model does rune use?
No built-in model — you configure your own via environment variables: Anthropic (ANTHROPIC_API_KEY), OpenAI or compatible API (OPENAI_API_KEY, RUNE_LLM_MODEL), or local via Ollama (RUNE_LLM_BASE_URL, RUNE_LLM_MODEL). Without a configured model, the server outputs facts only.