Purpose
go-docs-mcp is an MCP (Model Context Protocol) server written in Go that lets AI access documents in multiple formats. It handles PDF, TXT, Markdown, DOCX, CSV, and images through a single binary, avoiding the need to deploy a separate server for each format.
Main Capabilities
- Multi-format support: PDF, TXT, MD, CSV, DOCX, and images.
- 13 tools in total, covering document listing, format and dependency status queries, text reading, search, summarization, metadata, outlines, table extraction, image extraction, URL download, forced OCR, and image OCR.
- OCR: automatically falls back to tesseract for scanned documents; OCR can also be forced on PDFs or images.
- Caching: mtime-based in-memory cache to avoid repeated extraction.
- Security: directory locking to prevent path traversal; read-only operations; URL downloads limited to 50MB with Content-Type validation.
- Portable: supports macOS and Linux.
Usage
Prerequisites: Go 1.25+; poppler for PDF; tesseract for OCR (optional); pandoc for DOCX (optional). TXT, MD, and CSV require no additional dependencies.
Configure the document directory via the DOCS_MCP_DIR environment variable; after placing documents in that directory, the server will discover them automatically.
It can be integrated with Claude Code, Claude Desktop, or any MCP client. The server communicates via stdio using JSON-RPC 2.0.
Use Cases
Scenarios where AI needs to uniformly read documents in multiple formats, perform OCR on scanned files, extract tables or images from documents, or retrieve document content from a URL.
Limitations
- Only supports macOS and Linux.
- Read-only, no write operations.
- Can only access files within the DOCS_MCP_DIR directory.
- URL downloads are limited to a maximum of 50MB.
- Image extraction returns at most 10 images per call.
- Some formats depend on external tools (poppler, tesseract, pandoc).
Project Information
Licensed under the MIT License, Copyright 2026 Drolosoft.
