MCP Lens connects to Model Context Protocol servers, fetches tool definitions, and runs a multi-layer evaluation to surface protocol violations, quality issues, security risks, and real-world usability problems.
Requires Python 3.11+, Node.js 18+, and uv
A comprehensive toolkit for evaluating MCP server tools across protocol compliance, quality, security, and LLM-assisted usability.
Add, edit, delete, and test MCP server connections from the web UI. Supports multiple auth flows including OAuth 2.0 and DCR.
Fetch and browse tool definitions with full schema detail, parameter tables, and raw JSON schemas. Upload tools manually when servers are unreachable.
35+ checks across protocol compliance, tool quality, security analysis, and LLM-assisted usability evaluation.
Test tools from an AI agent's perspective. Auto-generates scenarios and validates tool selection, argument generation, and disambiguation.
Detects annotation mismatches, prompt injection surfaces, SQL injection risks, and data exfiltration patterns in tool definitions.
Run evaluations headlessly in CI pipelines or programmatically via 28 REST API endpoints. Export reports as JSON, YAML, or PDF.
Weighted scoring by severity with configurable pass/fail thresholds. Critical failures in protocol or security force gate failure.
Compare evaluation reports over time to catch regressions. Track false positives with justifications across sessions.
Server configs, encrypted auth tokens, and evaluation reports stored locally. Optional PostgreSQL support for production deployments.
Static analysis plus LLM-assisted testing for comprehensive coverage.
Missing names, invalid schemas, malformed annotations, schema validation against the MCP specification.
Vague descriptions, missing parameter docs, naming inconsistencies, tool overlap detection.
Annotation mismatches, prompt injection surfaces, SQL injection risks, data exfiltration patterns.
Description clarity, tool selection, argument generation, overlap disambiguation, safety resistance.
Run evaluations with any supported LLM provider. Compare results across models.