Agent Reliability MCP server

What it does

A hosted MCP server that exposes a curated knowledge graph about testing, benchmarking, and auditing autonomous AI agents. Each claim returned is bound to registered sources with citations and confidence values, designed for developers building or evaluating agent systems.

• Query best practices for distinguishing real evals from benchmark contamination when testing autonomous agents.
• Look up technical details about LLM-as-judge calibration, fault injection testing methods, and approval gate frameworks.
• Retrieve structured answers with sources and confidence scores to ground agent decision-making in verified evidence rather than unsourced assertions.

Tools this server exposes

Tool	Does
get_overview	Returns corpus overview including object counts, published topics, freshness, and what types of knowledge are available.
search	Full-text search over the knowledge graph with relevance scoring and field matching.
answer	Answers a question from the corpus, returning matched claims with sources and confidence values.
get_entity	Fetches a single knowledge object by id, including its claims and source citations.
get_topic	Lists knowledge objects tagged with a specific topic.
get_related	Returns graph neighbours of an object, showing outgoing and incoming relations with relation types.
get_sources	Returns the full source registry or sources cited by a specific object.
get_latest	Returns most recently verified knowledge objects as a freshness signal.

Before you enable it

An MCP server runs with whatever access you give it. Start with a development account rather than production, and check which tools the server exposes before you let an agent call them unattended.