goldensuite-mcp

# goldensuite-mcp > One MCP server exposing **every** Golden Suite tool — `goldenmatch`, `goldencheck`, `goldenflow`, `goldenpipe`, `infermap`, `goldenanalysis` — under a single endpoint. ```bash pip install goldensuite-mcp goldensuite-mcp serve --transport http --port 8300 ``` Or via container: ```bash docker run -p 8300:8300 ghcr.io/benseverndev-oss/goldensuite-mcp:latest ``` ## What it does `goldensuite-mcp` imports each sub-package's MCP tool list and dispatcher, composes them into a single `mcp.server.Server` instance, and serves them over stdio or Streamable HTTP. You point your MCP client at one endpoint and get the full Golden Suite — entity resolution, data quality scanning, transforms, pipeline orchestration, schema mapping, and trend analysis & regression detection. ## Tool collisions Tool names register on a **first-wins** basis. The registration order is: 1. **goldenmatch** — entity resolution (headline package; its tools win collisions) 2. **goldencheck** — data quality scanning 3. **goldenflow** — transforms & standardizers 4. **goldenpipe** — pipeline orchestrator 5. **infermap** — schema mapping 6. **goldenanalysis** — trend analysis & regression detection (registered last) If two packages register a tool with the same name, the later one is shadowed. **Shadowed tools are logged at WARNING level when the server starts**, so you can see exactly what happened: ``` WARNING goldensuite_mcp.server: tool collision: 'profile' from goldenflow shadowed by earlier goldencheck (first-wins) ``` If you need a shadowed tool, use that package's standalone MCP server instead (e.g. `goldenflow mcp-serve`). ## Curated tool listing (`GOLDENSUITE_MCP_TOOLS`) The full suite is ~105 tools. That many in one flat namespace makes an LLM's tool-selection noticeably worse, so **`list_tools` returns a curated headline set (~25 tools) by default** — the primary verbs of each package. Every other tool stays **fully callable by exact name**; the filter only trims …