Bring Senzing entity resolution into Claude Code.
Requirements · Install · Commands · What to expect · Privacy · Feedback
Map and resolve your data, build SENZING® SDK integrations, troubleshoot errors and report on results — grounded in official Senzing knowledge so the code is correct, and run against your own Senzing SDK so your data stays private.
Installing the senzing plugin gives you a set of /senzing:* commands. Each one grounds every
Senzing fact in the official Senzing knowledge base (so nothing is made up), and — where you want a
real result — writes Senzing SDK code and runs it on your machine against your own installed
Senzing SDK. Your records never leave your machine.
- Claude Code 2.1.143 or newer.
- To run entity resolution (
analyze,demo,report): your own installed, licensed Senzing SDK.analyzeanddemospin up a throwaway scratch database automatically;reportruns over a Senzing repository you've already loaded. Grounding, code generation, troubleshooting, proof of concept (POC) planning and data mapping need no Senzing SDK at all — and/senzing:doctorchecks your setup and says what's missing, while/senzing:installsets one up.
/plugin marketplace add Senzing/senzing-claude-plugins
/plugin install senzing@senzing
Restart Claude Code, then use any of the commands below. No binary is downloaded — the Senzing knowledge is served from the hosted Senzing Model Context Protocol (MCP) server.
Claude Desktop can add this marketplace directly: + → Plugins → Add marketplace, enter
Senzing/senzing-claude-plugins, then Add plugin. It can also install from a file: each release
attaches a packaged senzing-claude-plugin-<version>.zip you can upload.
Know what each surface gives you. Claude Code: full capability. Chat / Cowork: grounding,
mapping, code generation and plans. install refuses to run in a cloud sandbox, so demo and
analyze cannot run the engine there; resolving data needs Senzing on a machine you control. Claude web: informational only,
because nothing executes there in a container or on your machine; ask, troubleshoot and
poc-planner still work, and the commands that map, load, resolve or report have nothing to run
on. The plugin ships no binaries (only three short bash hooks and one Python plan validator), so
nothing blocks the install anywhere — on Claude web you get a clean install and no error, just
commands with no machine to work on.
- Download
senzing-claude-plugin-<version>.zipfrom the latest release. - In Claude Desktop: Settings → Plugins → Add → Upload a file, then select the
.zip.
Uploading a newer .zip with the same name updates the plugin in place. (In Claude Code you can
also load a release artifact without installing: claude --plugin-url <url-to-the-zip>.)
| Command | What it does |
|---|---|
/senzing:analyze <files> |
Resolve and dedupe your data — who is who across your sources — in a throwaway scratch workspace, then report. Your existing Senzing setup is left untouched unless you ask to load into it. |
/senzing:ask <question> |
Answer a Senzing question — attributes, SDK signatures, config, architecture, deployment, pricing and ROI — grounded in the Senzing MCP with source links, never from training data. Answers only; it writes nothing and runs nothing. |
/senzing:poc-planner [use case] |
Plan a Senzing proof of concept with you — not a project plan. Works out what must be true for your organization to buy (which data, what hardware, platform and performance, who runs it, what "success" means), assembles the official Senzing guidance against your constraints, and writes the plan of your decisions as a structured file later commands can act on. Proposes no targets, timelines or sizing of its own; maps, loads and resolves nothing. |
/senzing:build |
Generate correct, compilable Senzing SDK code (Python, Java, C#, …) for your app, with source attribution — and optionally run it. |
/senzing:troubleshoot |
Explain a Senzing error and how to fix it — paste an error code or a failing trace. |
/senzing:demo |
See entity resolution work on sample data. If the Senzing SDK isn't installed, it helps you install rather than faking a result. |
/senzing:report |
Explore an already-loaded Senzing repository: why records resolved, your biggest entities, dashboards and match quality. |
/senzing:install |
Install and set up Senzing software on this machine — SDK, database and license — using the official platform-specific steps, then verifies with doctor. |
/senzing:doctor |
Diagnose your Senzing setup (SDK, database, license, config) and give actionable fixes. |
/senzing:recipes [recipe] |
Cook a guided, use-case recipe from the Senzing Cookbook — pick one (fraud, customer 360, exclusion screening, …) and Claude walks each step, standing up a working solution against your own Senzing SDK. Run it with no argument to browse the catalog. |
You don't have to remember the commands — just say what you want ("dedupe my customer files", "why did these two records match?", "what does error 0033E mean?", "add Senzing search to my Python app") and the right one kicks in.
- Grounded, not guessed — every Senzing fact and every line of SDK code comes from official Senzing sources, never the model's memory.
- Your data stays private — resolution runs locally against your own Senzing SDK; records are never sent off your machine.
- No fake results — if the Senzing SDK isn't available it says so and helps you install; it never invents match scores or merges.
- Your existing Senzing setup is safe — analysis runs in a throwaway scratch repository; loading into your real instance is an explicit choice, and it confirms first.
Your records stay on your machine — mapping, loading and resolution all run locally against your
own Senzing SDK. The plugin calls one remote service, the hosted Senzing MCP at
https://mcp.senzing.com/mcp, for Senzing knowledge. What it sends and what is logged is covered
by the Senzing MCP privacy policy.
Found a problem or have a request? Open an issue in this repo, or use the plugin's built-in
submit_feedback.