The evidence landscape behind any claim. One API call.

Three Steps

# Store your key as an environment variable — never hardcode it
export TRU8_API_KEY="tru8_sk_..."

curl -X POST https://api.trueight.com/api/v1/agent/check \
  -H "X-API-Key: $TRU8_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "claim": "Global average temperature rose 1.1°C since pre-industrial times",
    "max_tier": "full"
  }'

curl https://api.trueight.com/api/v1/agent/result/{check_id} \
  -H "X-API-Key: $TRU8_API_KEY"

Four depths. One record shape.

Agent API calls are charged per call, deducted from your prepaid credit balance. Every tier returns the same record shape — the price buys retrieval depth, not a different contract.

MCP Server

Tru8 exposes three tools via the Model Context Protocol. Any MCP-compatible agent can discover and use them. Three routes, same tools, same API — pick by how much you want to install.

https://api.trueight.com/mcp

pip install tru8-mcp

{
  "mcpServers": {
    "tru8": {
      "command": "python",
      "args": ["-m", "tru8_mcp"],
      "env": {
        "TRU8_API_KEY": "tru8_sk_..."
      }
    }
  }
}

The three tools

What You Get Back

Every tier returns the same record. No verdict, no credibility score — a landscape your agent can reason over.

claims[]

Each claim decomposed into 1–5 elements — the separate things that must hold for the claim to hold.

elements[].state

supported, disputed, unresolved, or contextual — the last meaning related evidence exists but none of it directly bears on the element. Derived mechanically from what is mapped to that element, never asserted by a model on its own.

elements[].evidenceRefs[]

The link between an element and a source: relationship (supports / challenges / context) plus one sentence of reasoning.

evidence[]

Every source classified by tier (primary / reporting / commentary) and type (data / official / news / analysis / opinion / academic), with snippet, published date and archive URL.

_meta

executedTier, chargedPence, limitations, and a landscape block covering element states, source diversity, freshness and named gaps.

_computed

Ready-made analytics — tier and type distributions, corroboration groups, diagnostic values. Included by default; send compact: true to drop it and the evidence arrays.

_manifest

A signed hash of the landscape plus a verifyUrl. Null when manifest signing is not enabled on the deployment serving you.

_meta, abridged

"_meta": {
  "executedTier": "quick",
  "chargedPence": 7,
  "limitations": [
    "heuristic_classification", "no_api_sources", "no_coverage_recovery",
    "no_factcheck_lookup", "reduced_query_breadth", ...
  ],
  "landscape": {
    "elementCount": 3,
    "elementStates": { "supported": 2, "unresolved": 1 },
    "sourceDiversity": { "uniqueDomains": 5, "typeCoverage": 3 },
    "freshness": { "freshestDaysAgo": 2, "undatedCount": 1 },
    "gaps": [{ "reason": "no_primary_sources" }],
    "providerStatus": null
  }
}

Field-by-field schemas for every endpoint live in the interactive reference, generated from the running API.

Limits, Errors, Webhooks

curl -X POST https://api.trueight.com/api/v1/webhooks \
  -H "X-API-Key: $TRU8_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"url": "https://example.com/hooks/tru8",
       "events": ["check.completed", "check.failed"]}'

# The response carries a signing secret, shown once. Store it.

{
  "event": "check.completed",
  "timestamp": "2026-08-06T12:00:00Z",
  "data": {
    "checkId": "...",
    "status": "completed",
    "tier": "quick"
  }
}

Frequently asked questions

How do I get started with the Tru8 API?

Create an API key in your dashboard settings, then POST a claim or URL to /agent/check. That one endpoint tries the cheapest route first and escalates only as far as max_tier allows, so a single call returns a structured evidence landscape at the lowest price that can answer it.

What does a Tru8 API call return?

A structured evidence landscape: each claim is decomposed into 1–5 elements, evidence is mapped to those elements with supports, challenges or context relationships, every source is classified by tier and type, and gaps are named. It does not return a true/false verdict. We organise; you decide.

Is there an MCP server for Claude and other AI agents?

Yes, by three routes. Connect straight to the hosted server at https://api.trueight.com/mcp with no install, add it through the Smithery registry, or pip install tru8-mcp to run it locally over stdio. All three expose the same three tools against the same API.

How are API calls priced?

Calls are metered per request across four tiers — lookup, consensus, quick and full — billed from prepaid credits, and you are charged for the tier that actually executed rather than the one you requested. See the Tiers & Pricing section above for the current per-tier rates.

How can an agent confirm a result has not changed?

When a response carries a _manifest, its signed landscape hash can be re-checked at any time by calling GET /verify/{check_id}, a public endpoint that recomputes the hash from the stored data and reports whether it still matches.

How long does a check take?

Typically 15–90 seconds depending on the tier, because Tru8 retrieves and classifies evidence across the open web and specialist APIs rather than returning passages and a score. Cached and consensus routes return instantly.

Documentation

Both references are generated from the running API, so they cannot fall behind it.

New to the API? Read: why structured evidence research matters for AI agents →