v1.4.7 — MCP Resources & Prompts
What's New
MCP Resources (4)
Clients can now read live state from GoldenMatch without calling tools:
| URI | Description |
|---|---|
goldenmatch://dataset/stats |
Record count, cluster count, match rate, pair count |
goldenmatch://dataset/clusters |
Top 50 clusters by size with member IDs |
goldenmatch://config/current |
Active matchkeys, thresholds, scorers, blocking rules |
goldenmatch://dataset/schema |
Column names, types, and sample values |
Resources are contextual — only listed when data is loaded.
MCP Prompts (5)
Guided workflows that chain the right tools in the right order:
| Prompt | Description |
|---|---|
deduplicate-walkthrough |
Full ER pipeline: profile → configure → run → review → export. Optional focus param (accuracy/recall/balanced). |
investigate-cluster |
Deep-dive a cluster: view members → explain matches → fix bad merges → shatter if needed. |
compare-records |
Field-by-field pair comparison with score breakdown and plain-English verdict. |
data-quality-audit |
Profile data → scan issues → recommend cleaning → suggest transforms. |
pprl-setup |
Guide through privacy-preserving linkage: assess sensitivity → configure → run → compare. |
Why this matters
- Resources let MCP clients (Claude Desktop, Cursor, etc.) inspect GoldenMatch state passively — no tool calls needed for read-only queries.
- Prompts lower the barrier to complex workflows. Instead of knowing which of 30 tools to call and in what order, users select a prompt and the LLM follows the guided sequence.