Entity Enricher MCP Server

Entity Enricher MCP Server

A hosted Model Context Protocol server for Entity Enricher, available at https://entityenricher.ai/api/mcp/ (Streamable HTTP).

From an MCP-compatible client you can:

• Run any job on your client's own model (delegate_llm_calls=true): no platform model call, no credits.
• Design reusable schemas from sample data or documents and edit properties directly.
• Enrich single entities or lists, with multilingual fields and multiple models.
• Fuse results and recover failed expertise domains without repeating successful work.
• Resolve recurring objects to semantic identities and curate their aliases or uncertain matches.
• Derive relational tables and migrations for your PostgreSQL, MySQL or SQLite database.
• Benchmark enrichment, sample generation and schema generation on your own tasks.

Schema validation and model agreement do not establish factual truth or freshness. Inspect the actual sources, failures and partial outcomes. A successful generation, entity-layer admission and application on your external replica are distinct outcomes.

The MCP server needs no local installation. Optional database delivery uses ee-database on your replica host; its DSN stays there. Managed hosts can provision automatically; manual pairing is also supported.

Run jobs on your own model

Every start tool accepts delegate_llm_calls=true. The job then makes no platform model call and spends no credits: the model behind your MCP client answers every model call of the job, and the platform keeps building the prompts, enforcing the output schema, validating and normalizing each answer, saving the records and syncing the databases.

1. Start the job with delegate_llm_calls=true and no explicit model (omitted or auto, otherwise delegation_requires_auto). It pauses as soon as its first model call is ready.
2. Read llm_requests on get_job_status: each entry is one model call in MCP sampling/createMessage shape (system prompt, messages, tools, toolChoice) with a purpose naming the step and an expires_at deadline. Binary attachments are referenced as enricher://attachments/{id} resources, never inlined; extracted text is already in the messages.
3. Run the request on your own model and hand the result back with answer_llm_request. The platform validates it exactly as it validates a provider's answer; a rejected answer comes back as the same conversation with the error in a failed tool result, to fix and answer again.
4. Repeat until the job completes. Records are stamped with the model key mcp::client at zero cost; an unanswered request fails the job after fifteen minutes (llm_request_timeout).

The plan must include the feature (llm_delegation_not_in_plan). Protocol details and edge cases — concurrent requests, repair requests on enrichment, attachments, timeouts — are in the delegated LLM calls guide.

Quickstart

Option 1 — OAuth (recommended)

For claude.ai, Claude Code, Cursor, and any MCP client that implements the standard OAuth flow. No API key to create or paste — the client discovers the authorization server automatically, your browser opens the Entity Enricher consent screen, and the connection acts on your behalf with your own role. Revoke it anytime under Settings → API Keys → Connected Apps.

claude mcp add --transport http entity-enricher https://entityenricher.ai/api/mcp/

Then run /mcp in a session and pick Authenticate — your browser opens the consent page. More options (project .mcp.json, API-key fallback): examples/claude-code/

Settings → Connectors → Add custom connector with URL https://entityenricher.ai/api/mcp/, then click Authorize on the consent screen. Walkthrough: examples/claude-ai-remote.md

Register the URL with no headers and the client prompts you to sign in: examples/cursor/mcp.json

Option 2 — API key (static JSON configuration)

For clients configured via a JSON file rather than an interactive sign-in (Claude Desktop, Continue, Zed) — and for headless/CI use.

1. In the Entity Enricher web UI: Settings → API Keys → New organization access key. Pick a role — operator (read-mostly), editor (create/edit schemas), or owner (full control, required for benchmarks). Copy the ent_… value; it's only shown once.

2. For Claude Desktop, edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

   {
     "mcpServers": {
       "entity-enricher": {
         "url": "https://entityenricher.ai/api/mcp/",
         "headers": { "X-API-Key": "ent_your_key_here" }
       }
     }
   }
   

   Restart Claude Desktop. Full file: examples/claude-desktop/

Try it

List my Entity Enricher schemas, then enrich "Sanofi" against the pharmaceutical company schema in English and French.

The client discovers the tools, reads the selected schema and returns the result with a link. Automatic model selection is available; consequential choices are reviewed when needed.

Guides and tool descriptions

Server instructions explain the workflows. Tool descriptions explain individual calls, including preconditions, costs and consequential effects. Detailed modeling, recovery and migration guidance is loaded only when needed.

Use MCP resources/list to discover the guide index (enricher://docs) and each guide, then resources/read on the desired URI. Guides do not run a model. Clients decide how resources enter model context; reading them is not guaranteed to be token-free.

The following public recipes are generated from the same packaged Markdown the server serves. Edit the source guides in the main repository, not these generated copies.

If a client cannot read MCP resources, use these public links. The guides complement individual tool contracts; ordinary calls do not require reading them all.

Tools

59 tools, spanning the full schema-authoring and enrichment surface:

Tool signatures are the callable contract. The wrappers share backend services, but do not expose every REST/UI option. Schema mutations require editor; benchmark mutations/runs require owner plus a benchmark-enabled plan. Database registration/credentials require owner plus a sync-enabled plan. See each tool for its requirements.

Data resources

Jobs and errors

Long-running work uses this server's start → poll → fetch interface. Start tools return a job_id; generate_sample can already be paused or completed when it returns. Poll get_job_status, answer paused questions through answer_job_question, and retrieve records by list_records(job_id=...). A missing in-memory job is not proof of completion; check for persisted records. Cancellation does not undo earlier records or database writes.

A job started with delegate_llm_calls=true pauses on llm_requests instead of questions; see Run jobs on your own model.

Most failures return success: false, error_code and message; some older tools return only error or message. A successful MCP transport response does not imply successful work. Inspect classification warnings, failed model legs, and partial/rejected database outcomes even when an output is present. Detailed recovery is in the enrichment guide.

Scope and limitations

• The MCP single/batch enrichment tools do not expose web-search activation; sample generation does.
• Batch enrichment has no fixed 100-entity cap, but live quotas/credits can stop remaining work.
• The batch tool has no database_sync=false option; the single-entity tool does.
• Record deletion/restoration, detailed cost analytics, benchmark result import/export and some database administration remain in the web app or REST API.
• Schema documents use the supported JSON Schema dialect with Entity Enricher annotations; see the schema reference before authoring one directly.

Links

• Public MCP documentation
• REST API reference
• Client configurations
• MCP Registry — ai.entityenricher/enricher; manifest: server.json

About this repository

This public repository contains documentation and client examples. The server is embedded in the Entity Enricher backend and maintained in the private monorepo; this directory is synced as a git subtree. Licensed under the MIT License.