Talonic

What this does

Official Talonic MCP server. Lets AI agents extract structured, schema-validated data from any document (PDF, scan, image, DOCX, spreadsheet, form) via the Model Context Protocol.

When an agent needs structured data out of a messy document, raw OCR plus an LLM call produces unreliable results. With Talonic installed, the agent has a talonic_extract tool that returns schema-validated JSON with per-field confidence scores and a detected document type.

Tools (8)

• talonic_extract — Extract structured data from a document using a schema. Returns JSON with per-field confidence scores and a cost block (per-call credits, EUR, post-call balance) parsed from response headers. Supports drag-and-drop file_data, file_path, file_url, or document_id.
• talonic_search — Conceptual search across documents, fields, sources, and schemas in the workspace.
• talonic_filter — Filter documents by extracted field values (eq, gt, between, contains, is_empty, is_not_empty, etc.). Surfaces API warnings[] when a numeric operator is applied to a string-typed field, so the agent can suggest a schema-design fix instead of silently returning zero matches.
• talonic_get_document — Fetch full metadata for a single document.
• talonic_to_markdown — Get OCR-converted markdown for a document. Same cost block as talonic_extract on file paths; null on the document_id path.
• talonic_list_schemas — List saved extraction schemas. Returns both UUID and SCH-XXXXXXXX short id.
• talonic_save_schema — Save a schema definition to the workspace for reuse.
• talonic_get_balance — Read the workspace credit balance, EUR value, 30-day burn rate, projected runway, tier, and next-tier-reset timestamp. For budget-aware behaviour before large batches.

Plus two resources: talonic://schemas and talonic://webhooks/reference.

Install

Local (stdio, for IDE clients):

Works in Claude Desktop, Cursor, Cline, Continue, and Cowork.

{
  "mcpServers": {
    "talonic": {
      "command": "npx",
      "args": ["-y", "@talonic/mcp@latest"],
      "env": { "TALONIC_API_KEY": "tlnc_..." }
    }
  }
}

Hosted (Claude.ai connector):

Recommended path is OAuth — no API key in the config.

1. Open https://claude.ai/settings/connectors → "Add custom connector".
2. URL: https://mcp.talonic.com/mcp (the bare origin
https://mcp.talonic.com also works — the server routes both).
3. Click Connect. Sign in with Google, Microsoft, or SSO; approve the
consent screen.

Fallback (API key in URL):
https://mcp.talonic.com/mcp?apiKey=tlnc_your_key. Use this if you prefer
a static credential or cannot complete the OAuth consent flow.

The hosted endpoint advertises OAuth resource metadata at
/.well-known/oauth-protected-resource (RFC 9728); compliant MCP clients
and Inspectors discover the authorization server automatically.

Get an API key

Sign up at https://app.talonic.com. Free tier: 50 extractions/day, no
credit card. Settings → API Keys → Create New Key.

Not needed for the OAuth hosted install path above — Claude.ai handles
authentication via PKCE.

Links

- npm - https://www.npmjs.com/package/@talonic/mcp
- Repo: https://github.com/talonicdev/talonic-mcp
- Docs: https://talonic.com/docs/mcp
- Official MCP Registry: https://registry.modelcontextprotocol.io/v0/serv
ers?search=io.github.talonicdev/talonic-mcp
- Cursor Directory: https://cursor.directory/plugins/talonic
- Glama: https://glama.ai/mcp/servers/talonicdev/talonic-mcp
- Smithery: https://smithery.ai/servers/talonic/talonic