An MCP (Model Context Protocol) server that exposes Markdown documentation, API reference (C# XML-doc / TypeDoc), JSON Schema / OpenAPI specs, and JSON/JSONL data files to AI agents. From local folders, GitHub repositories, or straight from a service's OpenAPI URL.
A single server instance can host multiple libraries/frameworks side-by-side, each with its own sources. The AI picks which library to query via a library parameter, discoverable through the list_libraries tool.
| Direct file access | Docs MCP | |
|---|---|---|
| Security | Agent can read/write anywhere on the filesystem | Sandboxed per source with traversal protection |
| Discovery | Agent scans directories and reads files one-by-one | Index tools give instant overviews of every source |
| Search | Agent greps files manually, burning context | Dedicated search tools with regex or glob support |
| Large files | Entire file loaded into context | TOC + chapter extraction reads only needed sections |
| Data files | A large JSON export cannot be read at all | Structure, filtered rows and aggregates, streamed and capped |
| Multi-library | Agent must know every path/repo | One server, many libraries, self-describing |
| Source | Local files only | Local directories, GitHub URLs, or a live OpenAPI endpoint — no cloning or copying required |
| Live specs | Spec files copied by hand whenever they change | Fetched from the service and cached — still served when the service is down |
npm install
npm run build
# Single library
docs-mcpserver ./docs --name "My Docs"
# Multi-library
docs-mcpserver --config dev-docs.jsonSee sample-config.json for a complete multi-library example.
- Getting started — install, CLI, Claude Code / Claude Desktop integration
- Configuration — config file format, libraries, sources
- Tools — the tool groups the agent sees (docs / api / schema / data)
- Hosting — HTTP mode and IIS (httpPlatformHandler) setup
- Changelog — what changed per release
AGPL