Ground your AI agents in your verified commitments.

What is MCP?

The Model Context Protocol is an open standard developed by Anthropic that allows AI agents to query external data sources in a structured, schema-defined way, without needing custom integrations for each source.

Instead of scraping your website or relying on training data, an MCP-compatible agent can directly query your ClaimSentry endpoint and receive your official commitments as structured JSON. The agent then incorporates this data into its response, grounded in your verified facts.

Your ClaimSentry endpoint

Each ClaimSentry instance exposes an MCP server at:

https://<your-instance>/mcp/sse

The endpoint uses Server-Sent Events (SSE) transport. It is read-only and intentionally public, it exposes only the claims you explicitly mark as published in the dashboard. In production, protect it at the ingress level (nginx rate limiting, IP allowlist) rather than with authentication, so MCP clients don't need to manage credentials.

Claude Desktop configuration

Add the following to your claude_desktop_config.json file (usually at ~/Library/Application Support/Claude/ on macOS):

{
  "mcpServers": {
    "claimsentry": {
      "url": "https://<your-instance>/mcp/sse"
    }
  }
}

After saving, restart Claude Desktop. You will see a ClaimSentry tool available in the tool picker. Claude can now answer questions about your brand using your certified facts as context.

Cursor configuration

In Cursor, open Settings → MCP and add a new server entry:

{
  "name": "ClaimSentry",
  "transport": "sse",
  "url": "https://<your-instance>/mcp/sse"
}

Cursor will automatically discover the available tools and make them available to the AI assistant in your editor context.

Available tools

All three tools are read-only and only surface claims with status: published. Unpublished, draft, or proposal-status claims are never exposed.

[
  {
    "id": 42,
    "category": "sla",
    "claim_key": "uptime_99_9",
    "claim_text": "The service guarantees 99.9% monthly uptime.",
    "confidence": 0.95,
    "source": "ACME SLA v2.3"
  }
]

[
  {
    "id": 42,
    "category": "sla",
    "claim_key": "uptime_99_9",
    "claim_text": "The service guarantees 99.9% monthly uptime.",
    "confidence": 0.95,
    "source": "ACME SLA v2.3"
  }
]

{
  "id": 42,
  "category": "sla",
  "claim_key": "uptime_99_9",
  "claim_text": "The service guarantees 99.9% monthly uptime.",
  "verification_prompt": "Does the service guarantee 99.9% or higher monthly uptime?",
  "confidence": 0.95,
  "source": "ACME SLA v2.3",
  "published_at": "2025-06-01T14:32:00"
}

Response field reference

Category values

Authentication & security

The MCP endpoint has no authentication by design. Published claims are intentionally public, they represent facts you want AI agents to know about your brand.

For production deployments, restrict access at the ingress level:

# nginx example, rate limiting
limit_req_zone $binary_remote_addr zone=mcp:10m rate=30r/m;

location /mcp/ {
    limit_req zone=mcp burst=10 nodelay;
    proxy_pass http://backend:8000;
}

Enterprise customers can additionally restrict by IP allowlist and configure a dedicated subdomain for their MCP endpoint.

Verified commitments, sealed verdicts

The commitments you publish via MCP are the same ones ClaimSentry uses for challenge runs. When a challenge completes, each verdict is hashed (SHA-256) and submitted to an RFC 3161-compliant Timestamp Authority (TSA) the moment it is countersigned. The TSA returns a signed token (.tsr) that seals the verdict at that exact point in time — verifiable independently with standard tools such as openssl.

Under Article 41 of the eIDAS Regulation (EU 910/2014), an electronic timestamp is admissible as evidence in legal proceedings; issued by a qualified timestamping authority, it enjoys a presumption of the accuracy of its date and the integrity of the data. ClaimSentry supports qualified TSAs (dedicated configuration). The result is a closed loop: your official commitments are published via MCP, your agents are challenged against them, and every outcome is sealed into an Evidence Pack you can verify yourself — with no vendor lock-in.

Compatibility