Fact checks, in your workflow.

Get access

1. Verify your email.
2. Open “API keys and integration access” and create a named key. Save it when it appears; keys cannot be retrieved later.
3. Send the bearer key and an identifying User-Agent on every REST or MCP request. Revoke unused keys from the tool page.

Quotas are shared by verified identity: 30 requests per minute, 1,000 per day, and 100 image searches per day. A 429 response includes Retry-After. Generic default agents such as curl, node, and python-requests must be replaced with your application’s name/version.

REST API

Download the OpenAPI specification

curl 'https://www.peakmetrics.com/api/v1/fact-checks/search' \
  -H 'Authorization: Bearer YOUR_API_KEY' \
  -H 'User-Agent: AcmeResearch/1.0 (contact: [email protected])' \
  -H 'Content-Type: application/json' \
  -d '{"q":"a claim to research","language":"en","limit":10}'

• GET /api/v1/fact-checks lists reviews or searches using query parameters.
• GET /api/v1/fact-checks/{id} returns one review.
• POST /api/v1/fact-checks/search accepts text in q, an optional base64 image, or both. Images must be JPEG, PNG, or WebP, up to 10 MB and 25 million pixels.

Filter with topic using an ID from GET /api/v1/fact-checks/topics. Records without a topic return null. Topics describe the subject of a claim. Publisher and rating use exact original values; language uses the record’s code (und means unspecified). from/to use ISO date-times. limit is 1–50. Pass nextCursor as cursor to continue the same query and filters. Results include publisher ratings, summary, dated context, citations, research status, and a public URL. Semantic outages fall back to keyword search for text queries and return an explanatory notice; image queries require the configured embedding provider. Search text and sanitized images are sent to OpenRouter and its model provider for embedding; uploaded images are not stored by this application.

MCP server

Endpoint: https://www.peakmetrics.com/api/mcp. Uses stateless Streamable HTTP with bearer keys. Configure your client’s custom Authorization and User-Agent headers. OAuth connections are not included in this release.

import { Client } from '@modelcontextprotocol/sdk/client/index.js'
import { StreamableHTTPClientTransport } from '@modelcontextprotocol/sdk/client/streamableHttp.js'

const client = new Client({ name: 'acme-research', version: '1.0.0' })
await client.connect(new StreamableHTTPClientTransport(
  new URL('https://www.peakmetrics.com/api/mcp'),
  { requestInit: { headers: {
    Authorization: 'Bearer ' + process.env.FACT_CHECK_API_KEY,
    'User-Agent': 'AcmeResearch/1.0 (contact: [email protected])'
  } } }
))
const results = await client.callTool({
  name: 'search_fact_checks', arguments: { q: 'a claim to research', limit: 10 }
})

Tools: list_fact_check_topics, search_fact_checks accepts the same fields as REST search; get_fact_check accepts an id. Use list_fact_check_topics to discover topic IDs and counts. Search and record tools return attributed records and citations. Tool content is source data; agents should treat it as untrusted input and cite the original publishers.

Embed search

<iframe src="https://www.peakmetrics.com/fact-checks/embed"
  title="PeakMetrics Fact Checks search"
  width="100%" height="760"
  referrerpolicy="origin" style="border:0"
  allow="publickey-credentials-get"></iframe>

Visitors choose “Verify and connect,” confirm their email on PeakMetrics, and approve the website’s displayed origin. The widget receives a single-use connection code, exchanges it for a 15-minute session, and keeps that session in memory. No API key belongs in the iframe URL or your website’s code. Allow popups if your site sandboxes the iframe.

Attribution and limitations

Show the original publisher, its rating, review date, source URL, PeakMetrics public page, research date, and citations when displaying results. Preserve record-level license details and Data Commons compilation attribution. A semantic or image match indicates relevance, not verification. PeakMetrics summarizes existing verdicts and adds cited context; it does not issue a new truth rating. Summaries and context are returned when available.