Skip to content
DID.is Docs

Agents and tools

did.is/agents inspects the servers and agents your AI systems talk to. It has three tabs: MCP, A2A and Delegation (see Delegation receipts).

MCP servers

Section titled “MCP servers”

Enter an MCP endpoint URL (Streamable HTTP), for example https://mcp.example.com/mcp. The same check is also available as a standalone page with a README badge at did.is/tools/mcp-check (see README badge). DID.is:

  1. Tries the current protocol (2026-07-28, server/discover) and falls back to the 2025-11-25 session handshake.
  2. Lists all tools (paginated) and computes a SHA-256 fingerprint of each tool definition and input schema, using JSON canonicalization.
  3. Classifies each tool’s side effects by heuristic (read_only, write, network_egress, system_execution) and compares that with the server’s own annotations.
  4. Flags risk signals, such as hidden instructions in descriptions.
  5. Compares the tool inventory with DID.is’s previous observation of the same endpoint (drift).

Result fields you will see: status (INSPECTED, or AUTH_REQUIRED when the server answers 401; DID.is reports it and does not bypass it), negotiated version, server info, tools, inventory hash, drift and checks (transport, identity, protocol, auth, inventory, drift).

Limitations shown with every result:

  • Side-effect classes are DID.is heuristics over names, descriptions and schemas; declared annotations are the server’s own claims.
  • serverInfo is self-reported; MCP has no server identity attestation.
  • Drift is measured against DID.is’s previous observation of the same endpoint.

A2A agent cards

Section titled “A2A agent cards”

Enter an agent origin or card URL. DID.is fetches /.well-known/agent-card.json (falling back to the legacy agent.json), validates the A2A 1.0 shape (0.3 cards are recognised), checks interfaces and security schemes, and verifies signatures[] (JWS over the canonicalized card) with keys from a DID kid or an HTTPS jku.

Checks: fetch, schema, version, interfaces, origin, provider, security, signature.

A verified signature shows who signed the card. It does not show that the agent behaves as described.

From code: Agent trust APIs.

README badge

Section titled “README badge”

After a successful check on did.is/tools/mcp-check, copy the Markdown snippet shown under the result into your README:

[![DID.is MCP check](https://did.is/badge/mcp?endpoint=https%3A%2F%2Fmcp.example.com%2Fmcp)](https://did.is/tools/mcp-check?endpoint=https%3A%2F%2Fmcp.example.com%2Fmcp)

The badge shows DID.is’s last stored inspection of that endpoint: tool count, whether the tool definitions changed since the inspection before it, and the date, for example inspected · 7 tools · unchanged · 2026-10-20. It turns amber when definitions changed and reads not inspected yet for an endpoint DID.is has never checked.

Viewing the badge never re-inspects the server, so a README with heavy traffic does not send requests to your MCP server. To refresh it, run the check again. The image is cached for up to an hour.