@veracityapi/mcp

MCP server for VeracityAPI content trust scoring.

It exposes VeracityAPI's live HTTP API as thin MCP tools for Claude Desktop, Cursor, Windsurf, Continue, Goose, Hermes, and other MCP-compatible agents. The server does not implement scoring locally; VeracityAPI remains the source of truth for validation, billing, privacy handling, and scoring.

Tools

• analyze_text → POST /v1/analyze with type: "text"
  • Scores text for content trust, specificity/slop risk, weak provenance, evidence, recommended fixes, and workflow action.
• analyze_image → POST /v1/analyze with type: "image"
  • Scores an HTTPS image URL for visible synthetic-image artifact risk and workflow action.
• analyze_audio → POST /v1/analyze with type: "audio"
  • Scores a short HTTPS audio URL for synthetic-audio workflow triage with transcript return.
  • Optional caller transcript/context can be supplied.
  • The MCP server does not download, transcode, or store audio.
• check_balance → GET /v1/balance
  • Returns account credit balance and recent usage windows.

VeracityAPI results are workflow risk signals. They are not proof of authorship, truth, AI generation, voice cloning, speaker identity, or forensic determination.

Install / run

VERACITY_API_KEY=vap_... npx -y @veracityapi/mcp

Optional:

VERACITY_API_BASE_URL=https://api.veracityapi.com

Claude Desktop config

{
  "mcpServers": {
    "veracityapi": {
      "command": "npx",
      "args": ["-y", "@veracityapi/mcp"],
      "env": {
        "VERACITY_API_KEY": "vap_..."
      }
    }
  }
}

Cursor / Windsurf / Continue / Goose

Use the same stdio command:

npx -y @veracityapi/mcp

with environment:

VERACITY_API_KEY=vap_...

Tool examples

analyze_text

{
  "text": "Travelers should always be careful in tourist areas because scams can happen anywhere. Keep your belongings close and avoid strangers.",
  "context": { "format": "article", "intended_use": "publish", "domain": "travel safety" },
  "store_content": false
}

analyze_image

{
  "image_url": "https://veracityapi.com/demo/influencer-beauty-tonic.jpg",
  "context": { "format": "social_post", "intended_use": "publish", "domain": "influencer product post" },
  "store_content": false
}

analyze_audio

{
  "audio_url": "https://veracityapi.com/assets/demo-voice-message.mp3",
  "transcript": "Optional caller-supplied transcript or context.",
  "context": { "format": "social_post", "intended_use": "moderate", "domain": "voice-message authenticity triage" },
  "store_content": false
}

Privacy notes

• Text: when store_content=false, raw text is not stored in D1 analysis logs.
• Image/audio media storage: the MCP client sends store_content=false; VeracityAPI stores no media bytes, base64 payloads, or full media URLs, only URL hash plus hostname.
• Image: VeracityAPI stores no image bytes and logs only URL hash + hostname.
• Audio: VeracityAPI fetches capped HTTPS audio transiently, stores no audio bytes/base64/full URL, and logs only URL hash + hostname.
• Local MCP server only proxies JSON to VeracityAPI and returns JSON responses to the calling agent.

Errors

The MCP server maps VeracityAPI HTTP errors into readable tool errors:

• 401 missing/invalid API key
• 402 insufficient balance
• 429 rate limited
• 503 model unavailable

Run check_balance before autonomous loops.

Text auto-revise: set auto_revise: true on text requests to bill Analyze + revise at $0.010 / 1k characters and receive revised_text when recommended_action is revise. Evidence type values are strict enums for deterministic agent branching.