Senzing MCP Server — Documentation

Overview

The Senzing MCP server provides AI assistants with entity resolution capabilities through the Model Context Protocol. It offers 13 tools and 31 prompts for data mapping, SDK code generation, ER reporting and visualization, documentation search, error troubleshooting, and sample data access.

The server works entirely from pre-fetched documentation and indexed content. It never connects to live Senzing instances, never handles PII, and requires no authentication.

Setup

No authentication is required. If you use Claude, add it from the connector directory — it is the shortest path and there is nothing to paste. Everything below it is the manual route for any MCP client.

Claude Connectors Directory — easiest

The Senzing MCP server is published in Anthropic's Claude Connectors Directory at claude.ai/directory/senzing. Open that page, click Connect, and Claude wires up the endpoint for you — no configuration file, no copied URL.

Claude Code plugin — richest option

The Senzing plugin for Claude Code bundles this MCP server together with ten /senzing:* skills (analyze your own data, demo, build SDK code, doctor, install, troubleshoot, report, recipes, ask, PoC planning):

/plugin marketplace add Senzing/senzing-claude-plugins
/plugin install senzing@senzing

Source: github.com/Senzing/senzing-claude-plugins.

Claude Code — MCP server only

claude mcp add --transport http senzing https://mcp.senzing.com/mcp

Claude Desktop / Other MCP Clients

Add to your MCP configuration:

{
  "mcpServers": {
    "senzing": {
      "type": "http",
      "url": "https://mcp.senzing.com/mcp"
    }
  }
}

Network Access

Some environments — Claude Chat and Cowork among them — restrict outbound network access. For the full experience, ask whoever administers that environment to allow two domains:

Source URLs on senzing.com and github.com remain available for unrestricted environments; they carry provenance and, for CORD, serve the complete uncapped dataset.

Verify Connection

After connecting, ask your AI assistant: "What can this Senzing MCP server do?" — this calls get_capabilities and confirms the connection is working.

Tools

All tools are read-only except submit_feedback. Start any session by calling get_capabilities for up-to-date tool listings and suggested workflows.

Data Mapping

Documentation & Reference

SDK Development

Reporting & Analytics

Sample Data

Troubleshooting & Utility

Workflows

1. Map Source Data to Senzing Format

1. Call mapping_workflow with action='start' and your source file paths.
2. Walk through each step: Profile → Plan → Map → Codegen → QA.
3. Pass the state object from each response to the next call.
4. Run analyze_record to validate the output JSON and check feature distribution and coverage.

2. Generate SDK Integration Code

1. Call sdk_guide for platform-specific setup (e.g., topic='install', platform='docker').
2. Call generate_scaffold with target language and workflow (e.g., language='python', workflow='full_pipeline').
3. Call find_examples to find real-world usage patterns in 37 indexed repos.
4. Use search_docs for API details and deployment guidance.

3. Troubleshoot Errors

1. Call explain_error_code with the error code from your logs.
2. Follow the resolution steps in the response.
3. Call search_docs for additional context on the error class.

4. Evaluate Senzing

1. Use search_docs to learn about architecture and pricing.
2. Call get_sample_data for real test data from CORD datasets.
3. Call generate_scaffold with workflow='full_pipeline' for an end-to-end example.

5. Build Reporting & Dashboards

1. Call reporting_guide with your topic (e.g., topic='export', topic='dashboard', topic='data_mart').
2. Use search_docs for additional architecture context.
3. Call find_examples for real reporting implementations.

6. Migrate V3 to V4

1. Call get_sdk_reference with topic='migration' for all breaking changes.
2. Filter by module: topic='migration', filter='SzEngine'.
3. Use topic='flags' to understand the new flag system.

Troubleshooting

Connection Issues

If you cannot connect to the MCP server:

• Allow mcp.senzing.com. On Claude Chat and Cowork, outbound network access is restricted and this is the usual cause — the server cannot be reached at all until that domain is allowed. That one domain also serves the indexed GitHub example code (raw.githubusercontent.com is an optional alternative). See Network Access for what each domain covers.
• Verify the endpoint URL: https://mcp.senzing.com/mcp
• Ensure your MCP client supports Streamable HTTP transport.
• Check that outbound HTTPS (port 443) is not blocked by your network.
• Try calling get_capabilities to verify the connection.
• On Claude, adding the connector from claude.ai/directory/senzing avoids hand-typed endpoint mistakes entirely.

Empty or Unexpected Results

• search_docs returns no results: Try broader or different keywords. The server indexes Senzing-specific documentation, not general web content.
• find_examples returns no results: Try a more general query. Use language parameter to filter by programming language.
• explain_error_code not found: Verify the error code format. The tool accepts SENZ0005, 0005, or just 5.

Mapping Workflow Issues

• State errors: Always pass the complete state object from the previous response. The server is stateless — all workflow state lives in the client.
• Script download failures: Use download_resource as a fallback to get analyzer or entity spec files inline.

Data Privacy

Support

Email: support@senzing.com

Privacy Policy: mcp.senzing.com/privacy

Website: senzing.com