docs-mcp-server
A multi-tenant MCP server that indexes documentation from websites, git repositories, and local folders, and exposes them to AI clients via search and fetch tools.
README
docs-mcp-server
Multi-tenant Model Context Protocol (MCP) server for documentation search and retrieval.
Use one server to index docs from websites, git repositories, and local markdown folders, then expose them to AI clients through a clean MCP toolset.
Why this project exists
LLM answers are only as good as their context. docs-mcp-server makes your assistant query your real docs instead of relying on stale model memory.
- Run a single MCP endpoint for many documentation tenants.
- Keep sources fresh with scheduler-driven sync.
- Return ranked snippets and full documents with source URLs.
Who this is for
- Teams using MCP-compatible clients (Copilot, Claude, custom tools).
- Platform/dev experience teams curating internal + external docs.
- Engineers who want deterministic, inspectable doc retrieval.
Quick start (10–15 minutes)
git clone https://github.com/pankaj28843/docs-mcp-server.git
cd docs-mcp-server
uv sync
cp deployment.example.json deployment.json
uv run python deploy_multi_tenant.py --mode online
uv run python trigger_all_syncs.py --tenants drf --force
uv run python debug_multi_tenant.py --host localhost --port 42042 --tenant drf --test search
If search returns ranked results with URLs/snippets, your tenant is live.
Add to ~/.config/Code/User/mcp.json:
{
"servers": {
"TechDocs": {
"type": "http",
"url": "http://127.0.0.1:42042/mcp"
}
}
}
Documentation map
- Start here:
docs/index.md - First-time walkthrough:
docs/tutorials/getting-started.md - Operational recipes:
docs/how-to/ - API and config lookup:
docs/reference/ - Architecture and trade-offs:
docs/explanations/
Docs follow Divio quadrants: tutorials, how-to guides, reference, and explanations.
Runtime and architecture deep dives
- Runtime modes (online vs offline):
docs/explanations/runtime-modes-and-starlette.md - Mode evaluation checklist:
docs/how-to/evaluate-runtime-modes.md - Entrypoint and startup flow:
docs/reference/entrypoint-walkthrough.md - Core library rationale and mapping:
docs/reference/core-library-map.md
Core tools exposed via MCP
list_tenantsfind_tenantdescribe_tenantroot_searchroot_fetch
See full contracts in docs/reference/mcp-tools.md.
Source code orientation
- Process entrypoint:
src/docs_mcp_server/app.py - App composition:
src/docs_mcp_server/app_builder.py - Root MCP tool hub:
src/docs_mcp_server/root_hub.py - Tenant composition:
src/docs_mcp_server/tenant.py
Deep-dive guides that match these files:
docs/reference/entrypoint-walkthrough.mddocs/explanations/runtime-modes-and-starlette.mddocs/reference/mcp-tools.mddocs/reference/core-library-map.md
Contributing and quality gates
Development + CI validation loop:
uv run ruff format . && uv run ruff check --fix .
timeout 120 uv run pytest -m unit --cov=src/docs_mcp_server --cov-fail-under=95
timeout 120 uv run python integration_tests/ci_mcp_test.py
uv run mkdocs build --strict
More in docs/contributing.md.
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。
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。