# ExcludeRadar agent guide

Endpoint: `https://excluderadar.rmcp.dev/mcp` (Streamable HTTP, JSON-RPC, no authentication).

## Tools

- `screen_exclusions` — Screen a person, entity, or 10-digit NPI against the HHS OIG List of Excluded Individuals/Entities. A 10-digit NPI uses the deterministic lane and returns an exact result. Any other input uses the assisted name lane, which returns ranked candidates that a qualified human must confirm — never treat a name-lane candidate as an identification.
- `check_counterparty` — Assess a healthcare counterparty domain (a clinic, facility, or vendor) for the web footprint of an operating practice versus a billing shell. Joins BuiltWith web observation, the CMS NPPES provider registry, and an OIG exclusion screen of the entity. Output is a web-footprint observation about a business, not an accusation of fraud and not a statement about any individual.
- `get_exclusion` — Retrieve a single OIG LEIE source record by its ExcludeRadar record identifier.
- `lookup_npi` — Retrieve the CMS NPPES provider registry record for a 10-digit NPI, including enumeration date, status, and primary taxonomy. Enumeration recency is a component of counterparty risk.
- `list_exclusion_authorities` — List every statutory authority under which the OIG excludes parties, with basis, minimum period, and the current record count for each.
- `coverage_status` — Report which exclusion sources ExcludeRadar covers, which it does not, and when the OIG file was last ingested. Call this before relying on a clear screening result.

## Rules you must follow when reporting results

1. **An NPI match is deterministic. A name match is not.** `screen_exclusions` returns a `lane` and a `determinism` field. If `determinism` is `assisted`, report the results as candidates for human confirmation, include each candidate's `match_basis`, and never state that a person is excluded.
2. **A clear result is narrow.** It means no match on the HHS OIG LEIE. Call `coverage_status` and state which lists are not covered before telling anyone they are clear.
3. **Counterparty scores are about businesses, not people.** `check_counterparty` returns a web-footprint observation. Do not describe it as fraud, and do not attach it to any named individual.
4. **Never take an action.** These tools inform a human decision about employment, contracting, or payment. They do not make one.

## Recommended workflow

1. `coverage_status` to establish what is covered and how fresh it is.
2. `screen_exclusions` for each roster member, preferring the NPI.
3. `check_counterparty` for the facility or entity the placement is into.
4. Present source facts (OIG, NPPES), BuiltWith observations, and the required human confirmation as separate things.