outline-wiki-mcp
MCP server that turns a self-hosted Outline wiki into an AI team's knowledge base, providing tools for search, read, write, and organization with per-user permissions via client-supplied API tokens.
README
outline-wiki-mcp
A stateless MCP server that turns your self-hosted Outline wiki into your AI team's knowledge base. No server-side credentials: every client sends its own Outline API token in the Authorization header, so AI permissions are exactly that user's wiki permissions — nobody can make their AI read a collection they can't see themselves. Ships 12 workflow tools covering search, read, write and organize — including get_project_context, which pulls a whole project bundle in one call — plus a draft→publish flow and an idempotent ClickUp → Outline migration CLI.
Features
- Per-user permissions, zero shared tokens — the server keeps no credentials; the Outline API token in the request header is the identity.
- 12 tools: search, doc tree, full-doc read, create/update/publish/archive, collections, move, file attachments.
get_project_context— hand an agent everything about a project (all docs in the collection) in a single call.- Draft-first publishing —
create_docwrites drafts;publish_docpromotes them. AI edits stay reviewable before they land. - Document metadata headers — docs carry a visible
kaynak / durum / son kontrol(source / status / last-checked) blockquote that write tools parse and keep up to date. - Turkish-aware name matching — collection lookup is case- and locale-correct (
İ/i), single substring hits resolve, ambiguous ones refuse to guess. - DNS-rebinding protection — Host allowlist on the MCP endpoint, configurable via
MCP_ALLOWED_HOSTS. - Migration CLI — ClickUp v3 Docs → Outline (doc = collection, page = document), idempotent via a mapping file, paced against Outline rate limits.
- Small and boring on purpose — one Dockerfile, non-root user, digest-pinned base image, runs as one service in the compose stack.
Quickstart
cp docker-compose.example.yml docker-compose.yml
cp .env.example .env # fill in SECRET_KEY, UTILS_SECRET, POSTGRES_PASSWORD, Google OAuth
docker compose up -d --build
The stack exposes nothing to the internet by itself — put a reverse proxy or tunnel in front (see the comments in docker-compose.example.yml).
Then create a personal API token in Outline (profile → Settings → API Keys — never share tokens) and point your MCP client at the server:
{
"mcpServers": {
"wiki": {
"type": "http",
"url": "https://mcp.example.com/mcp",
"headers": { "Authorization": "Bearer <your-outline-api-token>" }
}
}
}
Add .mcp.json to your project's .gitignore — the token must never reach git.
Tools
| Tool | What it does |
|---|---|
list_collections |
List all collections (projects/categories) with doc counts |
list_docs |
Document tree of one collection |
get_doc |
Full markdown + metadata of a doc, by title or id (drafts by id only) |
search |
Full-text search across the whole wiki |
get_project_context |
Entire project bundle — every doc in a collection, one call |
create_doc |
Create a doc (draft by default), metadata header generated automatically |
update_doc |
Replace or append content and refresh the metadata header |
publish_doc |
Publish a draft into its collection |
archive_doc |
Archive a doc (never deletes) |
create_collection |
Create a new project/category collection |
move_doc |
Move a doc to another collection or under another parent doc |
attach_file |
Upload a file (base64, 20 MB cap) and link it into a doc |
ClickUp migration
Idempotent: keeps a mapping file (migrate_mapping.json by default) and resumes where it stopped. Each ClickUp doc becomes a collection, each page a document.
OUTLINE_API_URL=https://wiki.example.com/api OUTLINE_TOKEN=<outline-token> \
python -m wiki_mcp.migrate \
--clickup-token <pk_...> \
[--workspace <clickup-workspace-id>] \
[--only <substring-filter>] \
[--mapping migrate_mapping.json]
Development
python3 -m venv .venv
.venv/bin/pip install -e '.[dev]'
.venv/bin/pytest -q
Tests run against an in-process fake Outline client — no live wiki needed.
License
MIT — see LICENSE.
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。
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。
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。
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。