Verdigraph turns agent configs (Claude projects, OpenAI assistants, raw prompts) into deterministic, content-addressed "brain artifacts" with stable IDs you can pin in version control. The hosted MCP exposes brain build, verification, and evolution tools over HTTP, all backed by the same canonicalization pipeline that runs locally in pure Python. Twenty-five percent of revenue funds conservation. You'd reach for this when you need reproducible snapshots of agent state for audits, diffs, or automated testing, especially in multi-agent scenarios where one agent needs to inspect or route compute to another's cognitive graph. The source repo includes local-first tooling, so you can verify determinism without ever hitting the paid service.
claude mcp add --transport stdio io.github.jdhart81-verdigraph-mcp uvx verdigraph-mcp