AgentTrust
Score and stress-test your own MCP servers and agent workflows before you ship them to users or a marketplace.
Overview
AgentTrust is an MCP server for the Ship phase that scores AI agents and MCP servers on six axes and runs adversarial probes to verify quality before release.
What is this MCP server?
- Six-axis quality scoring for AI agents and MCP servers
- Adversarial probes to surface brittle or unsafe tool behavior
- LLM-judge evaluation via Groq or Cerebras free-tier keys
- MongoDB-backed persistence for audit history and reruns
- PyPI stdio package mcp-agenttrust (v0.1.2) for Claude Code-style hosts
- 6-axis scoring model
- Adversarial probe suite for agents and MCP servers
- Server version 0.1.2 on PyPI stdio transport
What problem does it solve?
You cannot tell if a custom MCP server will misbehave or leak trust until something breaks in a real agent session.
Who is it for?
Indie builders shipping custom MCP servers or multi-tool agents who want a structured security-and-quality gate with LLM-assisted judging.
Skip if: Teams that only need unit tests for a traditional REST API with no agent or MCP surface.
What do I get? / Deliverables
You get repeatable six-axis scores and probe results stored in MongoDB so you can fix issues before users or marketplaces see your tools.
- Six-axis quality scores for targeted agents or MCP servers
- Adversarial probe results highlighting failure modes
- Stored verification runs in MongoDB for comparison over time
Recommended MCP Servers
Journey fit
How it compares
MCP quality-verification server with adversarial probes, not a generic linter skill or hosting platform.
Common Questions / FAQ
Who is AgentTrust for?
Solo and small teams building MCP servers or agent toolchains who need measurable trust signals before launch.
When should I use AgentTrust?
Use it during Ship when an MCP server is feature-complete and you want scoring and adversarial checks before distribution or production agent access.
How do I add AgentTrust to my agent?
Install the PyPI package mcp-agenttrust, register the stdio server in your MCP host config, set GROQ_API_KEY or CEREBRAS_API_KEY plus MONGODB_URL, then invoke its verification tools from your agent.