Skip to main content
Glama

mcp-ai-accountability

Make your AI agents trustworthy and accountable.

This is a production-ready Model Context Protocol (MCP) server that helps AI agents (and the entrepreneurs who build them) measure, track, score, and improve reliability.

Why this exists

AI agents fail silently every day.
Tool calls time out. Schemas drift. Hallucinations happen.
Entrepreneurs lose money and trust because they cannot see why their agent is unreliable.

This MCP server turns reliability into a first-class tool that any agent can call.

Related MCP server: intent-engineering

What it does (simple language)

Imagine your AI agent is a delivery boy.
Sometimes he delivers the package late or to the wrong house.
This server is like a scorecard + GPS tracker + coach that:

  • Records every delivery (tool call)

  • Gives a reliability score (0-100)

  • Tells you why things went wrong

  • Suggests how to make the agent better next time

Perfect for founders who want accountable AI products.

Main Tools

Tool

What it does

record_interaction

Log a tool call result (success/fail + details)

get_reliability_score

Get current score for an agent or session

analyze_failures

Find patterns in failures

recommend_improvements

Get concrete suggestions to improve reliability

generate_audit_report

Full report for compliance / investors

check_mcp_health

Ping another MCP server and check if it is healthy

start_session / end_session

Track full agent runs

Quick Start

# Install
pip install mcp-ai-accountability

# Or run from source
uv sync
uv run python -m src.server

Add to your Claude / Cursor / any MCP client:

{
  "mcpServers": {
    "ai-accountability": {
      "command": "uv",
      "args": ["run", "python", "-m", "src.server"],
      "env": {
        "RELIABILITY_DB": "./reliability.db"
      }
    }
  }
}

For Entrepreneurs (non-developers)

You do not need to write code.
Just connect this MCP to your existing AI agent (Claude Desktop, Cursor, etc.).
Then ask your agent:

"Check my reliability score for the last 24 hours and tell me the top 3 problems."

The agent will use the tools automatically and give you a clear report.

Pricing Suggestion

  • Free tier: 1,000 interactions / month

  • Pro: $29/month – unlimited + audit reports + Slack alerts

  • Enterprise: custom (SSO + private hosting)

License

MIT

Built daily by the Prevalid team for the AI agent ecosystem.

Available Tools

8 tools
analyze_failuresA

Find the most common failure patterns for an agent. Returns the tools that fail most often and sample error messages.

Inputs: agent_id (required), hours, limit.

check_mcp_healthA

Simple health check against another MCP server (HTTP transport). Useful for multi-agent systems that depend on other MCP servers.

Inputs: server_url.

end_sessionB

End a tracking session and get a quick summary.

Inputs: session_id (required).

generate_audit_reportC

Generate a full audit-style report useful for investors, compliance or internal reviews.

Inputs: agent_id (required), hours.

get_reliability_scoreA

Get the current reliability score (0-100) for an agent. Score = percentage of successful tool calls in the last N hours.

Inputs: agent_id (required), hours.

recommend_improvementsC

Give concrete, actionable recommendations to improve reliability. Based on real failure data.

Inputs: agent_id (required), hours.

record_interactionB

Log one tool-call result. Call this after every important tool use so reliability can be measured.

Inputs: success (required), agent_id (required), tool_name (required), metadata, +3 more.

start_sessionA

Start a new tracking session for an agent. Call this at the beginning of a multi-step agent run. Returns a session_id that you should pass to later tools.

Inputs: agent_id (required).

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.6/5.0

    Scored across 8 tools

    Disambiguation5/5

    Each tool targets a distinct activity: session lifecycle (start/end), logging interactions, score retrieval, failure analysis, improvement recommendations, audit reports, and external health checks. No two tools overlap in a way that would cause confusion for an agent.

    Naming Consistency5/5

    All tools follow a consistent verb_noun pattern in snake_case: start_session, end_session, record_interaction, get_reliability_score, analyze_failures, recommend_improvements, generate_audit_report, check_mcp_health. The naming is predictable and uniform.

    Tool Count5/5

    8 tools is well-scoped for an accountability/reliability tracking server. Each tool covers a distinct part of the workflow (session management, interaction logging, scoring, analysis, recommendations, reporting, health checks) without redundancy or bloat.

    Completeness4/5

    The core lifecycle is covered: session start/end, interaction logging, reliability scoring, failure analysis, recommendations, and audit reports. Minor gaps exist such as session history or detailed interaction listing, but the existing tools form a complete workflow for the stated purpose.

    Maintenance

    ActivityInactive
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers