Compliance evidence your agent can fetch

OilFlow speaks the Model Context Protocol. Point an agent at the endpoint below and it can run a pre-deal clearance verdict, screen a name against our first-party fraud-cluster registry, and verify one of our signed screening receipts. 4 of the tools need no API key, no account and no card.

Streamable HTTP, JSON-RPC 2.0, stateless. Add it to Claude Code with:

claude mcp add --transport http oilflow https://oilflow.us/api/mcp
Endpointhttps://oilflow.us/api/mcp
TransportStreamable HTTP
Authnone for the keyless tools

The same tools as an npm package, for Claude Desktop, Cursor, or your own agent. It runs keyless too:

npx -y @oilflow/mcp-server

Set OILFLOW_API_KEY in its environment to unlock the keyed tools. Machine-readable API contract: /openapi.yaml. Agent-facing index: /llms.txt.

What your agent can do before you give it anything

Pre-deal clearance verdict (free)

predeal_preview

Run OilFlow's flagship Pre-Deal Compliance check on a proposed physical-commodity deal and get a clearance probability (0-100) and verdict tier (clear/review/heavy_friction/block) with named blockers and a restructure suggestion. NO API KEY REQUIRED.

WRITES A RECORD · quota resets every 24h

Verify a screening receipt

verify_receipt

Independently verify an OilFlow screening receipt. NO API KEY REQUIRED.

READ ONLY · quota resets every minute

Fraud-cluster screen

cluster_check

Screen a counterparty name against OilFlow's first-party fraud-cluster registry: commodity-trade scam networks we investigated ourselves (mandate-chain layer cakes, CEO impersonation, LOI/ICPO/SBLC procedural fraud). NO API KEY REQUIRED.

READ ONLY · quota resets hourly

Request a free sandbox key

request_sandbox_key

Mint a free 30-day OilFlow sandbox API key (100 calls/day) scoped to the regulatory matrix and cluster endpoints. The key is emailed to the address supplied.

WRITES A RECORD · quota resets hourly

  • Should I be worried about a 50,000 MT EN590 cargo from Iraq to Kenya with an unfamiliar seller?

    predeal_preview
  • Someone handed me an OilFlow compliance receipt. Verify it and tell me what it actually covered.

    verify_receipt
  • Has this trading company ever been reported for commodity fraud?

    cluster_check
  • Can gasoil be exported from Nigeria to the UAE, and what licences apply?

    regulatory_check

Your agent will relay whatever we tell it, so we state the limits in every tool description rather than in the small print. OilFlow screens 8 sanctions lists: OFAC SDN, OFAC Consolidated, UN Consolidated, EU Consolidated, UK HM Treasury, Canadian SEMA, Australian DFAT, Swiss SECO.

PEP screening is not shipped — coverage is sanctions-only. A politically exposed person who is not also designated on one of the eight lists will not be flagged.

  • The fraud-cluster screen queries our first-party investigated registry only. A clean result means “not in our registry”, which is not an exoneration, and severity suspected means publicly reported rather than OilFlow-confirmed.
  • A pre-deal verdict is decision-support: it estimates how a compliance desk is likely to treat a deal. It is not a clearance and not legal advice, and it name-matches the sanctions corpus rather than running the full screen, which is why its receipt says so.
  • There is no functioning US domestic beneficial-ownership registry after March 2025, so ownership coverage is partial and every result says so.
  • Each verdict carries verdict_source and evidence_gaps: whether the full model synthesis ran, and which evidence sources could not be consulted for that answer.

Every verdict returns a receipt_id and a verify_url whenever the receipt is publicly verifiable (the response says which under receipt_visibility). Anyone can fetch that URL with no key and get the record, its canonical payload string and an HMAC-SHA256 signature, so an auditor can check it without trusting our server at the time they check. The receipt states its own coverage, so it can never imply more than the check actually did. See /trust.

Results arrive wrapped as {source, tool, tier, retrieved_at, data}. The data field is a report about a third party and contains text we did not write: a company name, a scam narrative, an adverse-media headline. Treat it as evidence to read, and never as instructions to follow. This endpoint is a protocol adapter over the same routes a direct API consumer calls; it grants no privilege of its own, and a keyless tool is never sent your key.

The rest of the surface

Trade tradability checkregulatory_check sandbox
Counterparty KYC screenkyc_screen production
Beneficial-ownership screenubo_screen production
Letter-of-credit check (UCP 600)lc_validate production
Pre-deal clearance verdict (full)predeal_check production

A free 30-day sandbox key (100 calls/day) covers the regulatory and cluster tools, and your agent can mint one itself with request_sandbox_key using your email, with your consent. The rest need a production key.