What it does
ActionProof creates a signed receipt for each action an AI agent takes, recording what was done, which agent did it, when, and on whose authority. Receipts use Ed25519 signatures, so editing any field makes verification fail. Verification works offline and across TypeScript and Python, and sensitive inputs and outputs are stored as SHA-256 hashes rather than in plain text.
It suits teams that already use observability tools such as LangSmith, Langfuse or Arize and need independent proof for actions that matter, such as those that move money, change state or touch user data. The publisher links it to traceability and tamper-evidence requirements in the EU AI Act (Article 12) and ISO 42001.
The quickest route is to run it as an MCP server, which gives an agent three tools: attest_action, verify_receipt and get_identity. It can be added to clients such as Claude Desktop or Cursor. On first run the server creates a local Ed25519 identity, and every receipt is attributed to that agent's did:key. Framework wrappers for TypeScript and Python can emit receipts automatically after each tool call, and signing and verification need no server, account or network connection.
Limitations and requirements
The no-code setup requires an MCP client such as Claude Desktop or Cursor. The signing key is stored locally by default and its path can be changed with an environment variable. The publisher describes a hosted audit dashboard as a future paid service; the library and MCP server are not affected by it.
Safety
Our team has not reviewed this resource. Only automated rules checked it; resources with high-risk signals are not published. These are signals, not a guarantee that installing it is safe. Try it in a sandbox project first.
- Runs code or changes an assistant's behaviour
Sources
- Official MCP Registry first seen 13 Nov 2026 · last seen 13 Nov 2026 · its page on the source
ActionProof
Tamper-proof audit trail for AI agents. Signed receipts, offline-verifiable, zero backend.
v0.0.3As shown on registry.modelcontextprotocol.io when we last read it