clutter-mcp — synthetic data & synthetic document generator (MCP server)

A Model Context Protocol stdio server that lets an AI agent drive the whole Clutter pipeline — invent a believable synthetic company, mass-produce the documents, spreadsheets, emails, images and datasets it would really have, poll for completion, and fetch download URLs — through Clutter's public REST API, authenticated with an API key.

Use it to fill dev/test/demo systems (SharePoint, CRMs, file shares) with realistic content — synthetic documents and synthetic data that stand in for the real thing. Generate test data, demo data and training data, or test documents and training documents, on demand — or give an AI agent believable data to reason over, all without touching real or sensitive data.

It's a thin HTTPS client with no dependency on the rest of the Clutter codebase.

Quick start

1. Create a free account at https://clutter.run, open Settings, and mint an API key (clt_live_…, shown once).
2. Add the server to your MCP client config (Claude Code, Claude Desktop, etc.):

{
  "mcpServers": {
    "clutter": {
      "command": "npx",
      "args": ["-y", "clutter-mcp"],
      "env": {
        "CLUTTER_API_KEY": "clt_live_…"
      }
    }
  }
}

That's it — CLUTTER_API_URL defaults to https://clutter.run/api, so only the key is required.

Configuration

Run directly

CLUTTER_API_KEY=clt_live_… npx -y clutter-mcp

Typical agent flow

1. build_org — describe a company in a sentence → returns an orgId (omit projectId to auto-create a project). Free.
2. wait_for_org — block until the company build is ready.
3. create_run — generate content against the company:
   • doc_generator → a batch of documents (docx/pdf/xlsx/eml/jpg), folder-organised.
   • data_generator → one tabular dataset (xlsx/csv/json) with an exact row count.
   • doc_metadata_gen → one metadata record per document of a prior doc run.
4. wait_for_run — block until the run is complete.
5. list_run_documents + get_document_url, or build_zip + get_zip_url for the whole run as a single ZIP (folder tree preserved — ready to drop into SharePoint or a file share).

Tools

• Identity / usage: clutter_whoami, get_usage
• Projects: list_projects, create_project, delete_project
• Companies: list_orgs, build_org, get_org, wait_for_org, query_org, delete_org
• Runs: list_runs, create_run, get_run, wait_for_run, list_run_documents, get_document_url, build_zip, get_zip_url, delete_run

build_org and create_run are asynchronous (return an id immediately); use the wait_for_* tools to block until a terminal state, or poll get_org / get_run. Each tool maps to a REST endpoint and returns the raw JSON response.

Billing

Company builds, metadata and "ask the company" are free. New accounts get 10 free documents + 100 free data rows, then pay-as-you-go (all prices USD: $0.10/document, $0.06/10 data rows). A 402 from create_run means insufficient credit — top up at https://clutter.run/billing.

Loading content into SharePoint

Clutter hands you download URLs; your agent does the upload (via Microsoft Graph). Beyond a plain "drop files into a library", the generated metadata can drive automation — apply sensitivity labels, set permissions, assign retention, route flat files to the right site/library by metadata, or even generate the information architecture itself. Worked patterns with Graph calls: https://clutter.run/sharepoint-cookbook.md

Reference

Full REST API: https://clutter.run/api/docs (Swagger UI) · machine-readable guide: https://clutter.run/llms.txt

License

MIT