Developers

Quick start

Get a key
Access is by invitation while the method is being validated. Request access, then send the key with every request as Authorization: Bearer <key>.

Ask for a report

POST /v1/reports with a name, and a URL that identifies the subject if the name could be ambiguous.

Request a report

curl -X POST https://www.epistacklabs.com/v1/reports \
  -H "Authorization: Bearer $RAPI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"name": "Ørsted", "urls": ["https://orsted.com"]}'

A stored report comes back at once with status 200. Otherwise a job starts and you get 202 with its id.

Response 200: a stored report, abridged

{
  "report_id": "rep_…",
  "status": "complete",
  "generated_at": "…",
  "stale": false,
  "report": {
    "subject": { … },
    "method_version": "0.1.0+461aa947",
    "review_status": "machine-generated, unreviewed",
    "scores": {
      "factual": {
        "score": …,
        "interval_80": [ …, … ],
        "n_resolved": …,
        "confidence": { "level": "medium", "reasons": "…" }
      },
      "promise":    { … },
      "prediction": { … },
      "overall":    { … }
    },
    "featured_claims": [ … ],
    "caveats": [ … ]
  },
  "html_url": "/v1/reports/rep_…/html"
}

Response 202: a job has started

{
  "job_id": "job_…",
  "status": "queued",
  "poll": "/v1/reports/job_…"
}

3
Poll the job
GET /v1/reports/{id} with the job id until its status is complete. The finished report is in the same response.
Poll the job
```
curl https://www.epistacklabs.com/v1/reports/job_… \
  -H "Authorization: Bearer $RAPI_API_KEY"
```

Read every claim

GET /v1/subjects/{id}/claims returns each claim with its quote, source, fixed test, outcome and verified evidence.

Every claim behind the report

curl "https://www.epistacklabs.com/v1/subjects/<subject_id>/claims?claim_type=promise&page_size=50" \
  -H "Authorization: Bearer $RAPI_API_KEY"

One claim, abridged

{
  "claim_id": "…",
  "claim_type": "promise",
  "quote": "…",
  "quote_verified": true,
  "said_on": "…",
  "source": { "url": "…", "title": "…", "outlet": "…",
              "published": "…" },
  "resolution_criteria": "…",
  "outcome": "false",
  "on_time": false,
  "rationale": "…",
  "evidence": [
    { "url": "…", "publisher": "…", "published": "…",
      "quote": "…", "independent_of_subject": true }
  ],
  "judges_disagree": false,
  "responses": [ ],
  "review_status": "machine-generated, unreviewed"
}

Check a block of text. POST /v1/check takes a page or a draft, or a list of claims each with an optional source, and returns an outcome for each claim with no score. It runs as a job, and the text is deleted from the job when it ends.

Check a text

curl -X POST https://www.epistacklabs.com/v1/check \
  -H "Authorization: Bearer $RAPI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"text": "…", "author": "…", "said_on": "2026-10-01"}'

Send a reply. POST /v1/responses stores a reply or correction to a claim, with counter-evidence. It is shown on the claim as “response received, not yet reviewed”.

Reply to a claim

curl -X POST https://www.epistacklabs.com/v1/responses \
  -H "Authorization: Bearer $RAPI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"claim_id": "…", "name": "…", "role": "the subject",
       "statement": "…", "counter_evidence_url": "https://…"}'

Endpoints

POST/v1/reportsA subject’s report: stored at once, or a job (202).
GET/v1/reports/{id}A job’s status and, when complete, its report; or a stored report.
GET/v1/reports/{id}/htmlThe same report as a readable page.
GET/v1/subjects/{id}/claimsEvery claim, filterable by type, broad topic and outcome.
POST/v1/subjects/resolveFind a subject, or the people and organisations a text names.
POST/v1/checkOutcomes for the claims in a text, with no score.
POST/v1/responsesA reply to a claim, with counter-evidence.
GET/v1/methodologyThe method version and the published method.

MCP server

For assistants and research platforms. Three tools wrap the REST API, with no logic of their own.

lookup_reliability(name, topic?)
A subject’s report: scores for factual claims, promises and predictions, with ranges, confidence and featured claims. Call again later to collect a report that was still being made.
check_claims(text)
Outcomes and evidence for the claims in a text, with no overall score. Runs as a job of a few minutes.
get_evidence(subject_id, claim_id)
Everything behind one claim: the quote and source, the fixed test, the outcome with its reasoning, the evidence and any reply.

Run the MCP server

RAPI_API_URL=https://www.epistacklabs.com RAPI_API_KEY=rapi_… python -m rapi.mcp_server

Speed

A stored report: Under a second
A new subject: 10 to 17 minutesThen stored for the next caller
Stored reports: 90 daysThen served marked stale while a refresh runs

Limits

Requests: 60 a minutePer key
Report and check jobs: 20 a dayPer key
Request size: 256 KBText up to 50,000 characters

The review label

Every report and every claim carries "review_status": "machine-generated, unreviewed" until a person has reviewed it. If you show a verdict or a score, show the label with it.

Errors

401missing or invalid API keySend the key as a bearer token.
409name matches more than one subjectThe response lists the candidates; send subject_id.
413request body over the limitSend less text, or split it.
429rate limit or daily job limitWait and try again; stored reports are not affected by the job limit.
503daily spending limit reachedNew jobs resume the next day.

Full reference

The OpenAPI document lists every endpoint, field and response. It is at /openapi.json and readable with a key.

Request access How it works