# crossmatrix [Health: Active]

**Category:** 📋 Product Management  
**Repository:** https://github.com/praxec/crossmatrix  
**GitHub Stars:** 0  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/crossmatrix

## Description
Multidimensional House of Quality (QFD) MCP server — cross-dimensional relationship matrices.

## Tools
Capabilities this server exposes over MCP:

- **trace**
- **slice**
- **sourceRef.uri**

## Claude Desktop Quick Installation
No confirmed install details yet, so no config is published. Follow the setup steps in the repository above.

## Documentation

## What crossmatrix does

The crossmatrix MCP server represents a multidimensional House of Quality as linked dimensions, members, relations, observations, evidence, and derived views. It is suited to structured requirement analysis, trade-off review, and dependency tracing where a flat list of notes is insufficient. A model can connect needs, characteristics, files, features, or other project-defined members through typed relations and declared observation scales.

The server exposes two MCP tools. `crossmatrix.command` handles mutations such as registering dimensions, synchronizing members, declaring relations, recording observations, attaching evidence, and deprecating entities. `crossmatrix.query` provides reads and analysis, including status, descriptions, slices, traces, reports, exports, validation, coverage, conflicts, stale-source checks, and gap detection.

## How it works

Models are persisted as three linked document types under `.crossmatrix/`: configuration, dimensions, and state. An append-only JSONL audit file records each accepted mutation. Documents use stable `id@version` references, canonical JSON formatting, and atomic writes. Reads compute findings rather than storing them, while `check` validates documents and cross-references with path-specific diagnostics.

Commands use optimistic concurrency through `expectedVersion`, and `requestId` makes a mutation idempotent, including across restarts. Accepted changes receive audit entries with actor, request, version transition, and summary data. The `trace` and `slice` operations can start from a member ID or `sourceRef.uri`; pagination uses `limit` and `cursor`. A `contains` hierarchy can be rolled up so leaf observations contribute to an enclosing member.

The report query produces a Markdown House of Quality with weights, relationship matrices, a correlation roof, technical importance, targets, coverage, and findings. The stale query can compare stored source hashes with current `file://` sources and return affected members and cells.

## Setup and configuration

The crossmatrix MCP server can be run through the npm launcher with `npx -y @matthew-cochran/crossmatrix`, installed from a platform release asset, or built from source with Cargo. The npm package downloads and verifies a matching binary on first use, then runs it over stdio. A separately installed binary can be selected with `CROSSMATRIX_MCP_BINARY`.

The server locates its store using this order: `CROSSMATRIX_DIR`, the nearest ancestor containing `.crossmatrix/`, `$XDG_DATA_HOME/crossmatrix` or its default equivalent, and finally `./crossmatrix-state`. Run `crossmatrix-mcp init` to create a store and `crossmatrix-mcp check` to validate it. Release installers support Linux x86_64 and arm64, macOS x86_64 and Apple Silicon, and Windows x86_64 and arm64.

## Tools and capabilities

- Register dimensions and synchronize or propose members.
- Declare scales, relations, and contraction chains.
- Record observations using tokens, evidence, provenance, and source references.
- Query traces, slices, coverage, conflicts, findings, and stale references.
- Generate Markdown reports or export model, report, and findings data.
- Install, list, or remove bundled agent skills through the CLI.

## Limitations and notes

The crossmatrix MCP server communicates over stdio; the provided material does not describe an HTTP transport. Optional AI-assisted helpers are disabled by default, require Cargo feature flags and an explicit key, and are not included in release binaries unless built with those features. Their suggestions become proposed observations and do not affect analysis until confirmed.

Observation restrictions apply: numeric member weights require a deterministic import source, and measured values require observed provenance plus a source reference. Unsupported operating-system and CPU combinations are rejected by the installers. The legacy full-model command flow remains available for in-memory follow-up queries, but the documented persistent workflow uses the versioned store and audit log.

_Full upstream README: https://allmcps.com/mcp/crossmatrix/readme_