Rune MCP server

Rune

Ask questions about a codebase. Rune finds the evidence, your own AI model explains it, and every statement is checked against the code before you see it.

Install

Linux (x64) and macOS (Apple Silicon):

curl -fsSL https://raw.githubusercontent.com/thecolourfoundation/rune/main/install.sh | sh

Windows: download rune-windows-x64.exe from the latest release and run it from a terminal.

The installer verifies a SHA-256 checksum and puts rune in ~/.local/bin.

Use

cd your-project rune "how does routing work"

The first run scans the project (a few seconds). Rune prints ranked findings with file:line evidence, then your model writes a short explanation from that evidence. Any statement that cites evidence Rune didn't retrieve, or quotes text that isn't there, is dropped and counted.

rune explain shows the evidence trail behind any fact, and rune verify checks the stored understanding against the code as it is now.

Bring your own model

Rune has no model of its own. Point it at yours with environment variables:

# Anthropic (early support) export ANTHROPICAPIKEY=your-key

# OpenAI or any OpenAI-compatible API export OPENAIAPIKEY=your-key export RUNELLMMODEL=model-name

# Local model with Ollama (no key, nothing leaves your machine) export RUNELLMBASEURL=http://localhost:11434/v1 export RUNELLM_MODEL=model-name

If no model is configured, Rune prints the evidence, then setup help, and exits with code 2. Add --evidence-only to skip the model entirely. Set RUNELLMDEBUG=1 to see the model's raw reply and why any statement was dropped.

What leaves your machine

Only the evidence for your question goes to the model provider you configured: up to 40 short code snippets, never the whole project. Rune prints how many it is sending before it sends them. Use --evidence-only, or a local model, to send nothing.

Use it from an AI client (MCP)

Rune is an MCP server. Point any MCP client at it:

{"command": "rune", "args": ["serve", "/absolute/path/to/your-project"]}

The client gets tools such as runeagent, runesearch, runeexplain, runeverifyfact and runecheck_drift, with file, line and snippet citations. No key is needed here: the client's own model is the model.

Good to know

• Rune is read-only. It observes and reports; it never modifies your project.
• Best on JavaScript/TypeScript, shell, Lua, and config files. Other languages get thinner results for now.
• Explanation quality depends on your model. Small local models write plain, short explanations.
• Rune saves its scan in .rune/ inside your project. Add .rune/ to your .gitignore.
• Binaries are unsigned, so macOS or Windows may show a security warning. There is no prebuilt binary yet for Intel Macs or Linux on ARM.

Run rune --help for all commands. More detail: Docs.md.

License

MIT