MCP Server Diff

Know the moment an MCP server's contract changes. Each run probes the endpoint (handshake, capabilities, full tool inventory), diffs it against the previous snapshot, and reports exactly what moved: tools added, tools removed, tool descriptions or input schemas changed, server or protocol version bumps, and initialize-latency drift. Schedule it hourly or daily against the servers your agents depend on, and stop discovering breaking changes in production.

How it works

• First run stores a baseline snapshot (reported as baseline_created).
• Every later run probes fresh, compares by tool name and schema, updates the snapshot, and emits one row per change.
• State persists in a named key-value store on your account; the same URL can be tracked as separate streams with the stateKey input.

Output

Summary row:

{ "url": "https://example.com/mcp", "tools_count": 12, "tools_count_before": 11,
  "latency_delta_ms": 38, "change_count": 2, "note": "2 change(s) detected" }

Change rows:

{ "kind": "tool_added", "tool": "export_report", "detail": "Exports a report..." }
{ "kind": "tool_changed", "tool": "search", "detail": "changed: input_properties, required" }

Use cases

• Agent builders: schedule against every third-party MCP dependency; alert on change_count > 0.
• Server authors: outside-in release verification — did the deploy actually change what you meant to change, and nothing else?
• Platform teams: a contract-drift audit trail for the MCP surface you expose to internal agents.

Speaks streamable HTTP (JSON and SSE), protocol 2025-06-18, read-only — it initializes, lists, and leaves; it never calls the server's tools. Endpoints that are down, not MCP, or auth-walled without a token report cleanly and you are not charged; billing is per completed probe, baseline runs included.

Built by 1450 Enterprises — the team behind the MCP Server Probe, which this Actor extends into continuous monitoring. Its first scheduled target is our own public library server.