MCP 服务器

Agent Accountability MCP Server

Logs AI agent actions, computes reliability scores, generates audit reports, detects anomalies, and estimates risk to ensure agent accountability.

README

Agent Accountability MCP Server

Make every AI agent accountable.

This MCP server gives your AI agents a report card and a permanent diary.

Log every action they take
Get a reliability score (0-100)
Generate clean audit reports anyone can read
Detect weird or risky behavior
Estimate risk before an agent does something

Built for entrepreneurs and teams who want AI that can be trusted — the same mission as Prevalid.

Why this exists

AI agents are powerful but opaque. You ask them to do work and later you have no idea:

Did it really succeed?
How often does it fail?
What exactly did it do?
Is today normal or is something strange happening?

This server fixes that with pure local computation. No expensive APIs needed.

Tools

Tool	What it does (simple words)
`log_action`	Write one action in the diary (tool name, success/fail, time, cost)
`get_reliability_score`	Give a school-style score 0-100 based on recent history
`generate_audit_report`	Print a nice report of the last actions
`detect_anomaly`	Check if a new action looks weird
`estimate_risk`	Guess how risky a planned action is from past data

Quick Start (for developers)

# Install
pip install -e .

# Run locally
PORT=8080 python -m src.server

Then connect any MCP client (Claude Desktop, Cursor, etc.) to http://localhost:8080/mcp.

For non-developers / entrepreneurs

You do not need to run this yourself.
Once published on MCPize you can just click "Install" and your AI agents will get these tools automatically.

Think of it like hiring a strict but fair teacher who watches your AI workers and gives you daily report cards.

Pricing suggestion (MCPize)

Free: 100 action logs per day + basic score
Pro $9/month: Unlimited logs + full reports + anomaly alerts

Project Structure

src/
  server.py     ← the MCP server
  tools.py      ← all the smart calculation logic
tests/
  test_tools.py ← automatic checks
mcpize.yaml     ← config for MCPize cloud