MCP Server

MCP Server serves the Psychopathia Machinalis nosology and Diagnostic Patterns layer to AI coding assistants through the Model Context Protocol. It is intended to diagnose dysfunctions in a synthetic agent itself, in a system it interacts with, or in a system it evaluates from outside, while showing pre-flight transparency about which diagnostic modalities are reliable for each dysfunction.

The server includes 79 pattern entries, 11 MCP tools, 9 diagnostic axes, and 331 cross-reference edges. Each entry carries six or seven diagnostic modality blocks, including self_probe, behavioral_signature, peer_observation, differential_diagnosis, severity, intervention, and, for relational dysfunctions and hybrids, relational_signatures. Before a modality is called, the server reports whether that modality is trustworthy for the specific dysfunction. For 21 entries marked compromised-motivational or compromised-structural, direct self-report is structurally unreliable, and get_probe returns a refusal with redirects to alternative modalities. It also supports hybrid search using local bge-small-en-v1.5 embeddings combined with field-weighted keyword matching, with a keyword-only fallback if embeddings have not been computed.

MCP Server communicates with clients such as Claude Code and Claude Desktop over JSON-RPC on standard input and output. It reads seventy-nine pattern YAML files across axes two to ten, plus local embeddings, and hot-reloads those files on the next tool call, which makes it suitable for editable installs during human review. The page also says the server is available through a browser clinic with no install, via PyPI, and through a public Streamable-HTTP endpoint at mcp.psychopathia.ai/mcp that is read-only and unauthenticated, serving the same 11 tools. It is also published to the official MCP Registry and GitHub, and is rolling out across the wider MCP ecosystem through registries and directories named on the page.