mcp-name: io.github.CSOAI-ORG/agent-incident-reporter-mcp

Agent Incident Reporter MCP — the AI incident record a regulator can trust

[][] (https://proofof.ai/scoreboard) (https://csoai.org)

Your agent failed in production. The evidence is scattered stdout logs anyone can edit. When the EU AI Act Article 73 serious-incident report is due, "trust us" is not an answer. This MCP records every incident as a hash-chained, Ed25519-signed, tamper-evident record — editing any past record breaks every later one, visibly.

🚀 Quick start

pip install agent-incident-reporter-mcp
# Claude/any MCP client → add stdio server: python -m agent_incident_reporter_mcp

Tools

Why now

EU AI Act Article 73 requires providers of high-risk systems to report serious incidents. Article 50 transparency hits new generative systems on 2 August 2026. The market has runtime guardrails and assessment PDFs — nobody offers verifiable incident evidence. This does.

Pricing (canonical CSOAI ladder)

Free: unlimited local records (unsigned-dev). Pro £199/mo — Ed25519 signing + public verify URLs + the full governance suite (subscribe). Enterprise £1,499/mo · 48h bespoke assessment £4,950.

Part of the CSOAI fabric

290+ servers in the official MCP Registry (verified 2026-06-10) · csoai.org/fabric · CSOAI certifies; MEOK AI Labs builds. Records prove identity & integrity, not legal accuracy. MIT © CSOAI LTD (UK 16939677) · contact: nicholas@csoai.org

Configuration

Add to your claude_desktop_config.json (Claude Desktop) or your MCP client config:

{
  "mcpServers": {
    "agent-incident-reporter-mcp": {
      "command": "uvx",
      "args": ["agent-incident-reporter-mcp"]
    }
  }
}

Or: pip install agent-incident-reporter-mcp then run the agent-incident-reporter-mcp command (stdio transport).

Examples

Once configured, ask your assistant, for example:

• "Use the available tools to …"