SignalBrain

Lets an assistant emit, validate, and gate signed receipts for governed process changes, so autonomy is earned by re-runnable evidence.

What you can do

SignalBrain exposes three MCP tools: emit_receipt, validate_receipt, and gate_status. The agent writes spec-compliant claims about a change, checks that a receipt exists byte-for-byte on the merged ref, and reads its own earned-autonomy standing per change class. The underlying CLI re-runs a receipt's own measurement commands and turns the ledger into TRUST/GATE or per-class ELIGIBLE/GATE decisions.

When to use it

Use it when autonomous coding agents change systems that matter and you want a verifiable record of what was claimed and whether it held. It suits teams that want to move past approving every action forever and instead let a process earn reduced review per change class.

Requirements

Runs locally over stdio as a PyPI package, started with uvx --from "signalbrain[mcp]" sb-mcp. Python and uvx are needed. No authentication, environment variables, or headers are declared; the README states plain files, a CLI, and a GitHub Action, with nothing to host and nothing phoning home.