bookclaw-mcp
Exposes BookClaw's author workflow to MCP clients over Streamable HTTP, providing tools for managing books, projects, pipelines, and more.
README
bookclaw-mcp
A Model Context Protocol (MCP) server that exposes BookClaw's
author workflow to MCP clients over Streamable HTTP. It is a thin, stateless
client of BookClaw's REST API — see CLAUDE.md and
docs/superpowers/specs/2026-06-18-bookclaw-mcp-design.md.
Driving BookClaw from an LLM client? See docs/LLM-OPERATOR-GUIDE.md — a guide to hand to the LLM that will operate BookClaw through these tools (core concepts, the tool catalog, canonical workflows, and the confirmation-gate rules).
Setup
npm install
cp .env.example .env # fill in BOOKCLAW_AUTH_TOKEN and BOOKCLAW_MCP_TOKEN
npm start # listens on http://127.0.0.1:3849/mcp
BOOKCLAW_AUTH_TOKEN is BookClaw's own token (from BookClaw's .env).
BOOKCLAW_MCP_TOKEN is a token you choose to gate this server.
Install into Claude Code
This repo is also a Claude Code plugin marketplace. Quick HTTP install (server
running on :3849):
claude plugin marketplace add /home/paul/data/dev/bookclaw-mcp
claude plugin install bookclaw-mcp@bookclaw \
--config url=http://127.0.0.1:3850/mcp --config token=<BOOKCLAW_MCP_TOKEN>
For an always-on localhost service, run bash deploy/install-service.sh
(systemd user unit on 127.0.0.1:3850).
A stdio option (Claude launches the server itself) and full details are in
docs/INSTALL.md. The server also speaks stdio directly when
BOOKCLAW_MCP_TRANSPORT=stdio.
Tool profiles
The full surface is ~101 tools. Some MCP clients select tools less reliably past ~50–100, so the exposed set is configurable. The default exposes everything.
BOOKCLAW_MCP_PROFILE— a named bundle of tool groups. One of:all(default) — every tool.core— status, books, projects, chat, export, library.author— core + personas, series, craft.publishing— core + publishing, media, audiobook.marketing— core + marketing, website.
BOOKCLAW_MCP_GROUPS— optional override: an explicit comma-list of per-module groups (status,books,projects,chat,export,library,personas,series,craft,publishing,media,audiobook,marketing,website,escape-hatch). When set, it takes precedence over the profile.
The escape-hatch group (bookclaw_request, list_endpoints) is always
included, so every BookClaw endpoint stays reachable regardless of profile.
Unknown profile/group names fall back to all / are ignored, with a warning.
Tools
Core author loop: bookclaw_status, list_books, get_book,
create_book, set_active_book, get_book_files, read_book_file,
list_projects, get_project, create_project, create_pipeline,
advance_pipeline, get_project_files, chat, compile_project,
export_docx, list_library, get_library_entry.
Personas: list_personas, get_persona, create_persona,
generate_persona, update_persona, delete_persona, generate_persona_bio.
Series: list_series, create_series, update_series, set_series_refs,
get_series_worldbuilding, set_series_worldbuilding, add_book_to_series,
remove_book_from_series, set_series_reading_order, get_series_report,
get_series_divergence, delete_series. (The confirmation-gated series-asset
pull stays on the escape hatch.)
Craft analysis: list_beta_archetypes, run_beta_reader,
get_beta_reader_report, dialogue_audit, pacing_heatmap, craft_critique,
continuity_check, get_continuity_report, list_structures,
recommend_structure, check_outline_structure, structure_check,
get_plot_promises, extract_plot_promises, audit_plot_promises.
Publishing / export extras: export_blurb, export_project_blurb,
format_pro, get_manuscript_hub. (The multipart track-changes DOCX roundtrip
and cover typography stay on the escape hatch.)
Media: research, list_research_domains, set_research_domains,
generate_image, generate_book_cover, generate_cover_set,
generate_project_cover_set, list_cover_variants, list_image_providers,
generate_audio, list_voices, set_audio_config.
Audiobook prep: audiobook_cleanup, audiobook_pronunciation,
audiobook_ssml, audiobook_attribute. (Binary image/audio file-serving
endpoints stay on the escape hatch.)
Marketing / launch: list_launches, create_launch, get_launch,
update_launch, propose_launch_step, delete_launch,
propose_ams_campaigns, optimize_ams, draft_bookbub_ad, list_calendar,
create_calendar_event, plan_price_pulse, update_calendar_event,
delete_calendar_event, analyze_reader_intel, plan_translation.
Website: list_sites, get_site, create_site, update_site,
delete_site, add_site_book, remove_site_book, add_blog_post,
remove_blog_post, draft_blog_post, render_site. (Confirmation-gated site
deploy/publish, disclosure helpers, and the ICS calendar export stay on the
escape hatch.)
Escape hatch: list_endpoints, bookclaw_request.
Testing
npm test # unit + smoke
npx tsc --noEmit # type-check
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。