Let agents work with business documents.

13 tools on one hosted MCP server. Tag, extract, search, total, and index — without building an integration first.

The tools

13 tools, in three groups.

Names and descriptions come straight from the published manifest, so this page and the server never disagree.

An MCP host browsing Gemina tools by group with aggregate_documents selected and configured to sum totals by vendor

FileTag

3 tools

Tag, rename, and enrich any PDF or image. Free tier, no credit card.

  • files_create_uploadReserve a pre-signed PUT slot for a file the agent holds locally, then follow the returned next_tool_call recipe into tag_file. Bytes go straight from the agent host to storage and never traverse the LLM context.
  • tag_fileRun the FileTag pipeline against a previously uploaded slot. Returns extracted metadata, a suggested filename, six filename patterns, and a short-lived signed URL to an enriched copy with the metadata embedded in document properties or EXIF.
  • tag_urlFetch a public HTTPS URL server-side under strict SSRF guards and run the FileTag pipeline on it. Same result shape as tag_file; the bytes never traverse the LLM context.

Extraction

7 tools

Typed extraction over invoices, Hebrew documents, line items, raw OCR, and custom templates.

  • files_create_extraction_uploadReserve a pre-signed PUT slot for Core-OCR extraction. Distinct from the FileTag upload flow — follow the returned recipe into extract_document with one or more extraction_types.
  • extract_documentRun typed extraction on an uploaded slot. Choose extraction_types: ocr, invoice_headers, invoice_line_items, document_details_hebrew, document_line_items_hebrew, or custom_template. Asynchronous — returns the result directly or a correlationId to poll. Model selection and thinking, evaluation, correction, and coordinate toggles mirror the REST API.
  • get_extraction_resultPoll for the result of an asynchronous extract_document call using its correlationId. Returns the completed extraction, or an in-process status to poll again.
  • list_extractionsList past extractions, newest first. Filter by external_id, end_user_id, or an ISO date window, and paginate with skip and limit.
  • get_extractionFetch one extraction by id, including the full extracted data.
  • get_documentFetch one document by id, including all of its extractions.
  • submit_extraction_feedbackSubmit verified or corrected field values for a completed extraction and get back a per-field comparison summary. Each extraction accepts feedback once.

Document Intelligence

3 tools

Search the tenant's indexed documents and compute exact totals over them.

  • query_documentsSearch the tenant's indexed documents in structured mode (exact field filters), semantic mode (natural-language similarity), or hybrid mode (keyword and semantic fused with Reciprocal Rank Fusion). Returns matched documents with their extracted fields and scores.
  • aggregate_documentsCompute sums, averages, minimums, maximums, and counts over indexed documents, grouped by vendor, currency, document type, expense type, payment method, end user, month, or year. Money metrics are always split per currency unless a currency filter is given, so totals are never mixed.
  • index_documentManually (re)index one document into the searchable index — after corrections, or to backfill a document processed before indexing was enabled. Indexing otherwise happens automatically on every extraction once the tenant enables it.
Discovery

Look before you sign up.

A read-only mount answers tools/list with no credentials, so registries and clients can see what Gemina offers before anyone creates an account.

No API key requiredDISCOVERY
curl -X POST https://api.gemina.co/api/v1/mcp/public/ \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,
       "method":"tools/list","params":{}}'

Authenticated endpoint

Running a tool goes to https://api.gemina.co/api/v1/mcp/ with your API key as a bearer token or X-API-Key.

Machine-readable manifest

/.well-known/mcp.json describes the server, its auth, its file limits and every tool.

Works with any MCP host

Claude Desktop, Cursor, Claude Code, VS Code, Cline, Windsurf, OpenClaw and others.

A worked example

Documents in. A cited answer out.

Four steps an agent can run on its own.

A successful four-step agent run using extraction, query, aggregation and document tools to return an exact $12,450.75 total with three source citations
  1. 1

    Ingest

    Reserve an upload slot, put the bytes there, and run typed extraction. The document never passes through the model context.

    files_create_extraction_upload, extract_document
  2. 2

    Find

    Search the tenant’s indexed collection by field, by meaning, or both at once.

    query_documents
  3. 3

    Total

    Compute sums and counts in the database, grouped by vendor, currency, month, or type.

    aggregate_documents
  4. 4

    Cite

    Fetch the documents behind the numbers so the agent can show its working.

    get_document, get_extraction
Worth knowing

Chat is not one of the tools.

Agents get search, aggregation and indexing over MCP. Chat is not on that list. It runs over REST and the SDKs instead. A conversation carries state from one turn to the next, and an MCP tool call is stateless.

So an agent that wants a conversational answer composes one itself, from query_documents and aggregate_documents. That is usually what you want anyway. The agent keeps control of the reasoning.

Connect an agent in a minute.

1,500 free FileTag calls a month. No credit card.

Setup Guidegemina-mcp on GitHub