mycelium
Provides MCP tools for semantic search over personal knowledge sources using pluggable embeddings and local vector indexing.
README
mycelium
Shared toolkit for building agent plugins over personal knowledge sources (documents, notes,
and whatever comes next) — extracted out of the OpenClaw
plugins for paperless-ngx and Trilium Notes, which had independently grown near-identical
semantic search implementations. Those plugins — plus the 1Password secret-provider plugin —
now live in this same repo (apps/paperless-ngx, apps/trilium, apps/onepassword), so they
can consume the shared packages below as ordinary workspace dependencies instead of a
publish-then-consume round trip across separate repos. Each still publishes to npm
independently, under its own name.
Three problems, three packages:
| Package | Problem it solves |
|---|---|
@mycelium/embed |
"I want to use any LLM for embeddings/retrieval, not just one hardcoded provider." Pluggable client for any OpenAI-compatible endpoint (OpenAI, OpenRouter, Ollama, vLLM, LM Studio, ...), plus an opt-in local CPU fallback with zero API dependency. |
@mycelium/index |
"I want semantic search over my personal document/note corpus without running a vector DB server." A local, file-backed sqlite-vec index with incremental sync and hybrid lexical+semantic search (Reciprocal Rank Fusion). |
@mycelium/mcp |
"I want my plugin's tools usable outside one specific agent host." Bridges the same tool definitions onto a standalone MCP server, over stdio and Streamable HTTP. |
Plus @mycelium/tooling-config: shared biome/tsconfig presets, and
@mycelium/openapi-codegen: shared OpenAPI-schema-to-types codegen
logic used by the plugin apps' own generator scripts. Both published, but enablers rather than
features in their own right.
Why "mycelium"
The fungal network that links otherwise-separate organisms and moves nutrients between them — which is the job here: linking separate personal-data sources to any LLM provider, over any transport, without each plugin reinventing the same wiring.
Development
pnpm workspaces, TypeScript 7, Biome for lint/format, Changesets
for versioning and releasing the core/*/tools/* packages. Node only, deliberately — see
AGENTS.md for why this isn't Bun.
pnpm install
pnpm run build
pnpm run lint
pnpm run test
See AGENTS.md for repo layout and conventions.
License
MIT
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。