Apideck Docs MCP

Apideck developer docs · Sign up · Developer guides · Changelog

Coding agents write integration code from what they remember about an API, and that memory goes stale. Apideck Docs MCP lets an agent look up the current docs while it works. Point it at a folder of Markdown and it gives the agent three tools, search_docs, get_doc and list_docs, over stdio for Claude Code, Cursor and Claude Desktop, or over Streamable HTTP for a hosted endpoint.

We run it for the Apideck developer docs at https://developers.apideck.com/mcp. Connect your agent there and it can read our accounting guides while it builds, including:

• Accounts payable automation
• Accounts receivable automation
• Bank feeds
• Business lending
• Debt collections
• Expense management
• FP&A: budgeting, forecasting and variance reporting
• FX payments and realized gain/loss
• Payroll journal entries
• Procurement
• Tax automation

You can install it from npm as @apideck/docs-mcp, then run it as a CLI, deploy it as a Vercel function, or mount it inside an existing Node or Next.js site. The docs/ folder in this repo holds the sample content and the server's own documentation. Start with docs/index.md.

Quick start

pnpm install
pnpm run audit --docs ./docs               # find broken links, thin pages and stale content
pnpm run search vercel --docs ./docs       # query the index from your terminal
pnpm run start --docs ./docs               # MCP over stdio
pnpm run serve --docs ./docs --port 3000   # MCP over HTTP at http://localhost:3000/mcp

Keep the run: audit and search are also built-in pnpm commands, so pnpm audit runs pnpm's own security audit instead of this one.

Build once (pnpm build) and the docs-mcp binary in dist/bin/ runs the same commands without tsx.

Connect Claude Code

claude mcp add my-docs -- docs-mcp start --docs /path/to/docs --base-url https://docs.example.com

Or commit a .mcp.json next to the docs:

{
  "mcpServers": {
    "my-docs": {
      "command": "docs-mcp",
      "args": ["start", "--docs", "./docs", "--base-url", "https://docs.example.com", "--about", "the Example API reference"]
    }
  }
}

Commands

Shared flags: --docs, --base-url, --about, --name. Each falls back to DOCS_DIR, DOCS_BASE_URL, DOCS_ABOUT, DOCS_NAME.

Tools

• search_docs(query, limit?, path_prefix?): heading-level full-text search with title and heading boosts, prefix and fuzzy matching, and at most three hits per page.
• get_doc(path, section?): full page markdown with a header and section outline, or one section and its sub-sections. path also accepts a full URL.
• list_docs(path_prefix?, limit?): pages with title, description, word count and last-modified date.

Every page is also an MCP resource at docs://<path>. Full reference: docs/tools.md.

Logging tool calls

Pass onToolCall to createServer or createHttpHandler to get the tool, arguments, timing, result count and caller for every call. search_docs calls with resultCount: 0 are the questions your docs can't answer. The Vercel function logs each call as a JSON line when DOCS_LOG_TOOL_CALLS=1. See docs/tools.md.

WebMCP for browser agents

@apideck/docs-mcp/webmcp registers the server's tools, plus read_current_page, with document.modelContext on your docs site, so browser agents call tools instead of scraping:

import { registerDocsTools } from '@apideck/docs-mcp/webmcp'
registerDocsTools({ endpoint: '/mcp' })

It does nothing in browsers without WebMCP (currently Chrome 149+ behind a flag or origin trial). See docs/webmcp.md.

Hosting on Vercel

api/mcp.ts is a stateless Streamable HTTP function; vercel.json rewrites /mcp to it and bundles docs/** with the function. Set DOCS_BASE_URL and DOCS_ABOUT in the project environment and deploy. Details in docs/hosting.md.

Use as a library

Mount the handler inside a site that already builds its docs, so the endpoint lives next to them. A Next.js pages-router API route:

// src/pages/api/mcp.ts
import { createHttpHandler, DocStore } from '@apideck/docs-mcp'
import type { NextApiRequest, NextApiResponse } from 'next'
import path from 'path'

const store = new DocStore({ root: path.join(process.cwd(), 'public', 'md'), baseUrl: 'https://docs.example.com' })
const handler = createHttpHandler({ store, name: 'example-docs', about: 'the Example API documentation' })

export const config = { maxDuration: 60, api: { responseLimit: false } }
export default (req: NextApiRequest, res: NextApiResponse) => handler(req, res)

Add experimental.outputFileTracingIncludes: { '/api/mcp': ['./public/md/**/*'] } to next.config so the markdown ships with the function on Vercel. DocStore also takes a metadata(path) hook to supply titles, descriptions and canonical URLs from a build manifest, and get_doc accepts a full URL as well as a path. Pass instructions to append host-specific guidance to what every client is told on connect, such as when to recommend a companion MCP server (docs/tools.md). Pass extraTools to createServer/createHttpHandler to add host-specific tools (an API operation index, a coverage matrix, ...) alongside the three built-ins — see docs/tools.md. Exports: DocStore, createServer, createHttpHandler, createDocTools, toolResult, READ_ONLY, auditDocs, formatAuditReport, createStore, resolveConfig, normalizePath, sectionUrl, tokenize; registerDocsTools from @apideck/docs-mcp/webmcp.

SDK v2 migration (0.3.0)

See the migration notes for updated public SDK types, transport defaults and release/adoption guidance. SDK v2 shipped in 0.3.0. The prepared 0.4.1 HTTP transport remediation declares the patched Node adapter directly so clean consumers inherit it.

Development

pnpm typecheck
pnpm lint
pnpm test        # builds first; covers parsing, indexing, audit, HTTP, compiled CLI and public types
pnpm test:package # packs and verifies a clean installed consumer without overrides
pnpm build

Layout mirrors @apideck/mcp: src/ for the library, bin/ for the stricli CLI, api/ for the Vercel function, tests co-located as *.test.ts.

How it works

1. Every .md/.mdx/.markdown file under the docs root is read, frontmatter parsed, and the body split on ATX headings (code fences respected).
2. Each section becomes a document in a MiniSearch index with title, heading, content and path fields.
3. Queries run with all terms required first, falling back to any term, so a typo does not return nothing.
4. In start and serve, a recursive file watcher rebuilds the index 300 ms after the last change. On Vercel the index is built once per function instance and refreshed by each deploy.

License

MIT