Docpick
Turn scanned or digital documents into structured JSON from your coding agent using a schema you define, with local OCR plus an LLM.
Overview
Docpick is an MCP server for the Build phase that schema-driven extracts documents with local OCR and an LLM and returns structured JSON.
What is this MCP server?
- Schema-driven extraction: you define the target JSON shape and the server maps document content into it
- Local OCR plus LLM on the docpick PyPI package (0.1.2) over stdio MCP transport
- Document in, structured JSON out—suited for forms, receipts, IDs, and mixed layouts
- stdio MCP server (identifier docpick) for Claude Code, Cursor, and other MCP hosts
- No mandatory cloud OCR API in the positioning—designed for local processing workflows
- MCP server version 0.1.2
- PyPI package identifier docpick with stdio transport
- Source repository QuartzUnit/docpick on GitHub
Community signal: 1 GitHub stars.
What problem does it solve?
Agents and indie apps drown in PDFs and scans where the data you need is trapped in unstructured pages instead of typed fields.
Who is it for?
Builders adding document upload or inbox flows who want MCP-callable extraction with defined JSON shapes and local-first processing.
Skip if: Teams that only need full-text search in repos with no document schemas, or who require a fully managed cloud document AI with SLAs and no local stack.
What do I get? / Deliverables
After you register docpick, your agent can pass a document plus schema and get consistent structured JSON you can validate, store, or ship in features.
- Structured JSON matching your supplied schema per document
- Agent-callable MCP extraction workflow over stdio
- Repeatable document-ingest path for build and operate iterations
Recommended MCP Servers
Journey fit
Document-to-JSON pipelines are usually built and wired into products during the build phase as an integration surface for agents and backends. Integrations is the canonical shelf because docpick is an MCP stdio server you register so agents call extraction tools against your schemas, not a standalone research bookmark.
How it compares
MCP document-extraction integration, not a general web-to-markdown scraper or semantic code search tool.
Common Questions / FAQ
Who is docpick for?
Solo and indie builders using Claude Code, Cursor, or Codex who need agents to turn PDFs and images into JSON that matches their own schemas.
When should I use docpick?
Use it when you are scoping or building features that ingest forms, invoices, or scans and you want structured output without rewriting parsers for every layout.
How do I add docpick to my agent?
Install the docpick package from PyPI (0.1.2), add an MCP stdio server entry with identifier docpick in your host config, restart the client, and call its extraction tools with your schema.