FactGuard
Developer Early Access

The verification layer
your AI pipeline is missing.

FactGuard's REST API and MCP server turn any claim into a structured verdict — confidence score, cited sources, and reasoning chain — in a single call. Provenance (C2PA) tells you where content came from; FactGuard tells you whether it's accurate.

🔍 Cited verdicts ⚖️ Non-partisan engine 🇪🇺 EU AI Act–ready 🤖 MCP server included
Request early access See the API

The developer fact-check market is
essentially empty.

Every major fact-checking API is either read-only (Google), enterprise-only (Factiverse), or academically positioned (ClaimBuster). No production-ready, self-serve, verdict-generating API exists. With EU AI Act Article 50 enforcement arriving in August 2026, demand for AI output verification is accelerating.

The verification gap

AI systems hallucinate. RAG pipelines propagate stale claims. Content pipelines have no grounding layer. FactGuard closes this loop with a cited verdict at the point of generation or publish.

EU AI Act tailwind

Article 50 (full enforcement Aug 2, 2026) mandates disclosure and verifiability of AI-generated content. FactGuard's API produces structured, cited verification records suitable for compliance logging.

Compliance-ready

MCP white space

The two fact-checking MCP repos on GitHub are hobby experiments — neither is production-ready. FactGuard's MCP server is the first credible option for AI assistants and agent frameworks that need real-time grounded verification.

One endpoint. One call. Cited verdict.

POST a claim, get back a structured JSON verdict with confidence score, reasoning, and cited sources. No setup beyond an API key.

Request
# Verify a claim — POST /v1/factcheck $ curl -s -X POST \ https://api.factguard.org/v1/factcheck \ -H "Content-Type: application/json" \ -H "x-api-key: fg_live_YOUR_API_KEY" \ -d '{ "claim": "Vaccines cause autism." }'
Response 200
{ "claim": "Vaccines cause autism.", "score": 0.03, "reasoning": "Multiple large-scale studies find no causal link. The original Wakefield study was retracted and its author struck off.", "sources": ["Lancet", "CDC", "WHO"], "biases": ["Misleading"], "model": "groq_llama8b", "cached": false, "quota": { "limit": 100, "remaining": 87, "resetsAt": "2026-06-08T00:00:00Z" } }

Score scale

0.00–0.24
False
0.25–0.49
Mostly false
0.50–0.69
Mixed
0.70–0.89
Mostly true
0.90–1.00
True

MCP server — one config line

Drop FactGuard into any MCP-compatible AI host (Claude, Cursor, Zed, Windsurf, Continue) and your agent can verify claims in real time — with cited sources returned in context.

// mcp.json (Claude Desktop / Cursor / etc.) { "factguard": { "command": "npx", "args": ["@factguard/mcp", "--key", "$FG_API_KEY"] } }
Early access

Endpoints

Method + PathAuthDescription
GET /v1/health None Health check — no quota consumed
POST /v1/factcheck x-api-key Verify a claim, returns scored verdict

Choose the right model for your use case

Free-tier keys use groq_llama8b — fast and zero-cost. Pro/Scale keys unlock larger, more nuanced models.

Model IDProviderTierNotes
groq_llama8b Groq / Llama 3.1 8B Free Default · fastest · free quota
flash_lite Gemini 2.5 Flash Lite Pro+ Structured output · fast
groq_qwen32b Groq / Qwen3 32B Pro+ Thinking model · nuanced claims
flash Gemini 2.5 Flash Pro+ Balanced speed/quality
flash_25_pro Gemini 2.5 Pro Scale Highest quality · slowest

Simple, honest pricing

Start free. Scale when you're ready.

★  Early access — pricing may change before GA
Free / Dev
$0
forever · no credit card
  • ✓ 100 verifications / day
  • ✓ groq_llama8b model
  • ✓ Cited sources + reasoning
  • ✓ 30-day claim cache (free hits)
  • ✓ REST API + MCP server
  • ✓ Pro/Scale models
  • ✓ Higher daily quota
  • ✓ SLA + priority routing
Request access
★ Most popular
Pro
$49
per month
$0.0098 per verification · beyond quota: $0.012 / call
  • ✓ 5,000 verifications / day
  • ✓ All models incl. Qwen 32B & Flash
  • ✓ Cited sources + reasoning
  • ✓ 30-day claim cache (free hits)
  • ✓ REST API + MCP server
  • ✓ Overage at $0.012 / call
  • ✓ Email support
  • ✓ SLA + priority routing
Request early access
Scale
$199
per month
$0.004 per verification · beyond quota: $0.008 / call
  • ✓ 50,000 verifications / day
  • ✓ All models incl. Gemini 2.5 Pro
  • ✓ Cited sources + reasoning
  • ✓ 30-day claim cache (free hits)
  • ✓ REST API + MCP server
  • ✓ Overage at $0.008 / call
  • ✓ Priority routing + 99.5% SLA
  • ✓ Dedicated support channel
Request access

Cache hits are always free — when a claim has been verified within the last 30 days, the result is served from the corpus cache and does not count against your daily quota. cached: true in the response confirms this.

Built for builders who care about accuracy

Always cited

Every verdict includes 1–3 source names. Not summaries, not paraphrases — named publications your users can verify themselves.

Sources required

Calibrated score

A continuous 0.0–1.0 confidence score, not a blunt boolean. Your application can set its own threshold for how uncertain is "too uncertain."

Graded, not binary

Non-partisan engine

The same evaluation standards apply to every claim regardless of political origin. Neutrality is an engineering decision, not an editorial policy.

Neutral by design

Two-layer cache

Authoritative corpus from established fact-checkers (no TTL) sits above a 30-day LLM cache. Cache hits are free and typically sub-100 ms.

Fast + free cached

Server-to-server only

CORS is disabled. Keys are stored as SHA-256 hashes — plaintext never persisted. No stack traces or provider internals in error responses.

Secure by default

Reasoning chain

The reasoning field explains the verdict in plain language — suitable for logging, audit trails, or surfacing to end users.

No black boxes

Get your API key
before public launch.

Early-access developers get first access to keys, lock in early-access pricing, and have input on the roadmap — including batch endpoints, webhook callbacks, and async long-form verification.

We'll only email you when early access opens. No marketing.

I'm interested in:

By submitting you agree to our Privacy Policy. Unsubscribe anytime.

Want to see the verification engine before requesting access? The FactGuard web app runs the same engine — free, no account needed.

Try the web app free Read the methodology