Integration runbook
Teams building financial agents on LangChain, a home-grown gateway, or an enterprise agent platform.
Make every number an agent uses traceable and checkable — and make it say so when it does not know.
Each step below says which surface carries it and whether it works today. Every "Works today" step was probed against the live API on 2026-08-21; the rest say what is missing. Nothing here is written in the present tense because it is planned.
The endpoint is https://mcp.twmarketdata.com/mcp — the path matters, the bare host 404s. Any MCP-capable framework can connect, and the server enumerates its own tools on connect (24 today, covering queries, research, filings search and inclusion proofs).
The tool layer takes as_of, so an agent asking a historical question does not see the future. Disclosure datasets are filtered by disclosure date.
coverage.missing lists exactly what was requested and not returned, with a reason. Feed it to the agent as grounds for refusal: an empty data array beside a populated missing list is a complete and correct answer to 'what do you have', not an invitation to fill the gap.
twmd_verify_proof.py uses the Python standard library only. It imports nothing of ours and installs nothing from PyPI. Download it from /twmd_verify_proof.py and run it in your environment — a verifier that needed our code would be asking you to trust our code first.
One data access returns one signed receipt binding as_of, the query and the rows returned; stored as an attachment in LangSmith or Langfuse it becomes the data-side evidence for that decision.
There is no publicly callable receipt endpoint — every address probed does not exist. What the engine has today is v0, not v1, and until it is a callable interface this is not written as served.
CrewAI and AutoGen wrappers, plus a per-framework guide to embedding receipts in traces.
Not served.
Receipts and proofs speak to which data was used, at what point in time, whether it was altered, and how far coverage extended. They do not establish that the agent's conclusion is correct — that belongs to the model and reasoning layer, which no data vendor can vouch for.
For how the proofs map onto published standards, see Standards & Interop.
