Featured

Deploy OpenClaw in 60 seconds — 20% off logoDeploy OpenClaw in 60 seconds — 20% off

Launch OpenClaw on Hostinger in about 60 seconds and keep your agent live 24/7. Our referral link gives you 20% off, no coupon code needed.

Launch on Hostinger
Run your Hermes agent on Hostinger, fully managed logoRun your Hermes agent on Hostinger, fully managed

Launch Hermes on Hostinger in one click, fully managed, no VPS knowledge needed. Use code ZACAARON10 for 10% off.

Launch on Hostinger
Crawl and scrape any site into clean data, 10% off logoCrawl and scrape any site into clean data, 10% off

Firecrawl crawls and scrapes any site into clean markdown for your agent. Get 1,000 free credits, and new users get 10% off their first purchase.

Try Firecrawl free
6,000+ web scrapers for your AI agent, start free logo6,000+ web scrapers for your AI agent, start free

Apify gives your agent live web data: 6,000+ prebuilt scrapers and actors, MCP-ready. Sign up free with $5 in usage credits.

Try Apify free
One API to scrape, enrich, and extract the internet. logoOne API to scrape, enrich, and extract the internet.

Context.dev gives your agents a single API to scrape, enrich, and extract live web data — no proxies, no parsers, no maintenance.

Start building free
SetupClaw: done-for-you OpenClaw for founders & exec teams logoSetupClaw: done-for-you OpenClaw for founders & exec teams

White-glove OpenClaw for founders and exec teams (4–50+ employees): we install, harden, integrate your tools, and maintain it — secured from day one.

Get it set up for you
SEO data APIs for your agent, $1 free credit logoSEO data APIs for your agent, $1 free credit

DataForSEO gives your agent live access to SERP results, keyword data, backlinks, and on-page SEO data through one API. New accounts get a $1 credit, good for up to 20,000 keyword or backlink lookups.

Try DataForSEO free
Reach 48,000+ AI builders

A flat monthly placement in front of developers actively installing AI tools. No lock-in, cancel anytime.

Advertise here

Works with

Claude CodeClaude DesktopCursorVS CodeClineCodex CLIOpenClaw+ any MCP client

Install to Claude Code

This server doesn't publish a one-line install command. Follow the setup in the source repository.

Summary

A Model Context Protocol server for GitBook that enables reading content and driving a change-request write workflow.

README.md

gitbook-mcp

A Model Context Protocol server for GitBook. It lets an MCP host (Claude Code, Claude Desktop, Cursor, …) read your GitBook content and drive a change-request write workflow, over stdio or streamable HTTP.

Built on the official @gitbook/api client and @modelcontextprotocol/sdk 1.x.

  • Full API coverage: ~345 tools across 11 config-gated toolsets. The default (core + change-requests) keeps the surface small; write and destructive tools are gated by read-only mode plus an opt-in.
  • Resources: pages addressable as gitbook://{spaceId}/{pageId}.
  • Two transports: stdio (local) and streamable HTTP (multi-session, bearer-gated).
  • Resilient: request timeouts, bounded retries with full-jitter backoff, rate-limit handling.
  • Safe: read-only mode, env-only token, secret redaction, localhost-bound HTTP.

Status: published as @hoyongjin/gitbook-mcp (v2.0.0), with the original 11 tools enabled by default. Covered by unit, MCP-protocol, and manifest-contract tests. The stdio transport suits local IDE/CLI use. The HTTP transport adds bearer auth, DNS-rebinding protection, a session cap with idle reaper, per-IP rate limiting, health/readiness probes, and Prometheus metrics for hosted use.

---

⚠️ What GitBook's API can and cannot write

GitBook's public API has no direct "edit this page body" endpoint (createPage/updatePage exist only as the GitBook Assistant's _internal_ tool names, not as REST methods). Content authoring goes through:

  1. Change requests: open a draft (gitbook_create_change_request), review, then gitbook_merge_change_request to publish.
  2. Content import: gitbook_import_content imports a web page URL into a space/change-request/page, AI-enhanced by default. This is the supported "write content" primitive. Imports are asynchronous, and GitBook exposes no status-poll endpoint.
  3. Git Sync: for fine-grained, paragraph-level authoring, connect the space to a Git repo and edit markdown there. _(Out of scope for this server; it is the right path if you need precise prose edits.)_

So the server exposes many reads and a single coarse, review-gated write path. It cannot rewrite a specific paragraph through the API.

---

Install

# From source (local use):
cd gitbook-mcp && npm install && npm run build
# → run with:  node dist/index.js   (or `npm start`)

# Or, once published:
npx @hoyongjin/gitbook-mcp

The package is scoped to @hoyongjin/gitbook-mcp (the unscoped gitbook-mcp on npm belongs to a different project). Make sure your npm account owns the @hoyongjin scope before publishing.

Create a token at GitBook → Settings → Developer → Personal access tokens.

Configure in your MCP client

Claude Code

# Local build:
claude mcp add gitbook \
  --env GITBOOK_TOKEN=gb_api_xxx \
  -- node /absolute/path/to/gitbook-mcp/dist/index.js

.mcp.json / Claude Desktop config

{
  "mcpServers": {
    "gitbook": {
      "command": "node",
      "args": ["/absolute/path/to/gitbook-mcp/dist/index.js"],
      "env": { "GITBOOK_TOKEN": "gb_api_xxx" }
    }
  }
}

(After publishing, replace command/args with "command": "npx", "args": ["@hoyongjin/gitbook-mcp"].)

Run read-only (recommended unless you need writes) by adding "--read-only" to args, or "GITBOOK_READONLY": "true" to env.

Configuration

All configuration is via environment variables (see .env.example). The token is never taken from a CLI argument.

| Variable | Default | Description | | -------------------------------------- | ------------------------- | -------------------------------------------------------------------------------------------------------------------------------------- | | GITBOOK_TOKEN | — | Required. Personal access token (account-wide privileges). | | GITBOOK_ENDPOINT | https://api.gitbook.com | API base URL. | | GITBOOK_READONLY | false | Hide all write/destructive tools when true. | | GITBOOK_TOOLSETS | core,change-requests | Comma-separated toolsets to enable, or all. Unset/empty → the default two. Unknown key fails closed. See Toolsets. | | GITBOOK_ALLOW_DESTRUCTIVE | false | Opt-in to register new destructive ops (deletes, unpublish, admin removals). The legacy merge tool is exempt. | | GITBOOK_TRANSPORT | stdio | stdio or http. | | GITBOOK_HTTP_HOST | 127.0.0.1 | HTTP bind host. | | GITBOOK_HTTP_PORT | 3000 | HTTP port. | | GITBOOK_HTTP_AUTH_TOKEN | — | Bearer token required on the HTTP transport. | | GITBOOK_HTTP_MAX_SESSIONS | 256 | Concurrent-session cap; new initializes past it get 503 (1–100000). | | GITBOOK_HTTP_SESSION_TTL_MS | 300000 | Idle-session reaper TTL (1 000–86 400 000). | | GITBOOK_HTTP_SESSION_MAX_LIFETIME_MS | 3600000 | Absolute session lifetime; reaped regardless of activity. | | GITBOOK_HTTP_TRUST_PROXY | false | Trust X-Forwarded-* (enable only behind a trusted reverse proxy). | | GITBOOK_HTTP_ALLOWED_HOSTS | — | Comma-separated hosts appended to the DNS-rebinding allow-list. Required for proxied/non-loopback access (e.g. mcp.example.com). | | GITBOOK_HTTP_ALLOWED_ORIGINS | — | Comma-separated origins appended to the allow-list (e.g. https://mcp.example.com). | | GITBOOK_HTTP_RATE_LIMIT_WINDOW_MS | 60000 | Rate-limit window per client IP. | | GITBOOK_HTTP_RATE_LIMIT_MAX | 120 | Max /mcp requests per window per IP; 0 disables. | | GITBOOK_LOG_LEVEL | info | debug/info/warn/error (logs → stderr). | | GITBOOK_TIMEOUT_MS | 30000 | Per-request timeout (1 000–120 000). | | GITBOOK_MAX_RETRIES | 5 | Max retries on 429/5xx/transport errors (0–10). | | GITBOOK_MAX_CONCURRENCY | 8 | Max concurrent outbound GitBook HTTP requests per instance (1–256). |

CLI flags: --stdio / --http (transport), --read-only, --toolsets <csv|all>, --allow-destructive, --port <n>.

Tools

The server covers the full GitBook API (~345 tools), organized into toolsets you enable with GITBOOK_TOOLSETS. By default only core + change-requests are on (~89 tools), so tools/list stays small and tool selection stays accurate. Set GITBOOK_TOOLSETS=all for everything, or list the ones you need (e.g. GITBOOK_TOOLSETS=core,change-requests,sites).

Every tool is one API endpoint with its own input schema. Read tools are always available, write tools need !readOnly, and new destructive tools (deletes, unpublish, admin removals) also need GITBOOK_ALLOW_DESTRUCTIVE=true.

| Toolset | Default | Scope | | ----------------- | :-----: | ------------------------------------------------------------------------- | | core | ✅ | Identity, orgs/spaces/pages reads, search, content + import_content. | | change-requests | ✅ | Full CR loop: create, list/diff, reviews, reviewers, comments, merge. | | collections | | Collections + collection permissions. | | sites | | Sites, structure, customization, redirects, share links, permissions. | | site-ai | | Ask AI, agent settings, questions, glossary-adjacent AI. | | site-insights | | Analytics/insights, context connections, channels, site MCP servers. | | integrations | | Integration install/configure/token + space integrations. | | data | | OpenAPI specs, translations, glossary, custom fonts, storage. | | git | | Git Sync: import/export, provider installations, repo/branch listing. | | public | | URL→content resolution, subdomains, custom hostnames. | | org-admin | | Org members, teams, invites, SSO, permissions (highest privilege). |

The 11 original tools keep their exact names and live in core / change-requests, so existing setups are unchanged. Per-tool parameters are delivered live via each tool's MCP description + input schema.

Excluded carve-outs: 5 streaming (SSE) AI endpoints, 2 ephemeral PDF-URL generators, and a few trivial auth/heartbeat ops are not exposed as tools (the non-streaming AI equivalents are).

List tools (_list_, search) accept a body with limit/page and return nextCursor when more results exist. Typical write flow: create_change_requestimport_content (target that CR) → review/diff → merge_change_request.

Resources

Pages are exposed as resources at gitbook://{spaceId}/{pageId} (rendered as markdown). Enumeration is intentionally not provided; discover page ids via gitbook_list_pages / gitbook_search, then read by URI.

Transports

  • stdio (default): one server per process; the token stays in-process. Best

for local IDE/CLI integrations.

  • streamable HTTP (--http): one MCP session per mcp-session-id. Binds to

127.0.0.1 and enables DNS-rebinding protection (Host/Origin allow-lists). When GITBOOK_HTTP_AUTH_TOKEN is set, every request needs a constant-time-compared bearer token. Binding to a non-loopback host without an auth token is refused (fail-closed). Sessions are capped and idle-reaped; the /mcp endpoint is per-IP rate-limited. Operational endpoints: GET /healthz /livez /readyz (unauthenticated probes) and GET /metrics (Prometheus; bearer-gated when an auth token is set). Behind a TLS-terminating reverse proxy, set GITBOOK_HTTP_TRUST_PROXY=true and GITBOOK_HTTP_ALLOWED_HOSTS=<your-public-host>: DNS-rebinding protection matches the exact Host header, so a proxied or non-loopback deployment (including the Docker image, which binds 0.0.0.0) must add its public host(s) or every request is rejected with 403 Invalid Host. A Dockerfile is provided.

GITBOOK_TOKEN=… GITBOOK_HTTP_AUTH_TOKEN=… gitbook-mcp --http --port 3000
# → POST/GET/DELETE http://127.0.0.1:3000/mcp   ·   GET /healthz   ·   GET /metrics

Development

npm install
npm run typecheck   # tsc on src + test
npm run lint        # eslint
npm run format      # prettier --write   (format:check verifies)
npm test            # vitest: unit + protocol + manifest-contract tests
npm run generate:manifest  # regenerate src/tools/manifest.ts from @gitbook/api (CI checks drift)
npm run test:coverage  # + v8 coverage with an 80% gate
npm run build       # tsc → dist/ (+ chmod the bin)

CI (.github/workflows/ci.yml) runs npm audit → typecheck → lint → format:check → build → coverage on Node 20 & 22, then asserts a clean git diff. CodeQL and Dependabot run separately.

Architecture

src/
  index.ts            launcher: load config → pick transport → graceful shutdown
  config.ts           zod-validated env + CLI overrides (frozen Config)
  logger.ts           structured stderr logging + secret redaction + correlation
  request-context.ts  AsyncLocalStorage requestId/tool (log correlation)
  metrics.ts          in-process Prometheus registry (served at /metrics)
  limiter.ts          async concurrency semaphore (outbound rate control)
  server.ts           createServer(): McpServer factory (tools + resources)
  resources.ts        gitbook://{spaceId}/{pageId}
  version.ts          stable SERVER_NAME + runtime SERVER_VERSION (from package.json)
  gitbook/
    client.ts         typed wrapper over @gitbook/api (cursor pagination → {items,nextCursor})
    resilient-fetch.ts timeouts + retry/backoff + concurrency cap; injected as the client's fetch
    import-url.ts     URL guard for server-side-fetch ops (http(s) only, no credentials)
    errors.ts         HTTP/transport error classification → safe messages
  tools/
    manifest.ts       GENERATED tool surface (one row per API op; source of truth)
    toolsets.ts       toolset keys + defaults
    index.ts          manifest-driven registration under toolset/read-only/destructive gates
    factory.ts        generic handler factory (every non-bespoke op)
    bespoke.ts        the 11 grandfathered tools (hand-written handlers)
    shared.ts         ToolContext + result/guard helpers + annotation presets
  transports/
    stdio.ts http.ts  the two transports (http: sessions, rate limit, health, metrics)

Notes / gotchas (pinned to @gitbook/api 0.183.0)

  • tsconfig uses moduleResolution: "Bundler", not NodeNext. @gitbook/api

0.183 ships a .d.ts that re-exports with extensionless relative specifiers (export * from './client'); under NodeNext those don't resolve and the entire client surface (and the GitBookAPI namespace members) silently disappears. Bundler resolution accepts them, giving full typing with no facade. Internal imports use explicit .js extensions so the emitted ESM runs on Node.

  • Import runs live under the singular client.org namespace, not the plural

client.orgs (which holds list/search). They are different runtime objects.

  • gitbook_get_page markdown vs document: format=markdown returns rendered

markdown; format=document returns GitBook's structured Document JSON.

  • Node ≥ 20 (global fetch, AbortSignal.any).

License

MIT

See related servers & alternatives →

Related MCP servers

Browse all →

Related guides

Hand-picked reading to help you choose and use AI & ML servers.