ExpertPack MCP

Name: ExpertPack MCP
Author: brianhearn

作者：brianhearn版本 0.7.0i★ –i在你的电脑上运行

注册名称io.github.brianhearn/ep-mcp

安装

概览

作者提供的说明（英文原文）

Serve ExpertPack knowledge packs over MCP with hybrid search, graph traversal and provenance.

PopNx MCP 的三大支柱

这三项永远不会合并成一个分数：意见、测量和人气是不同的东西。

用户评价i暂无评价评价功能将在后续阶段与账户功能一起开放
我们的基础检查i1 项中通过 1 项依据公开信息检查；尚未扫描代码
使用情况i尚未统计★ –

它能帮你做什么

我们还没有为它写通俗的介绍。下面是我们从注册中心读取到的事实。

连接方式: 在你的电脑上运行
传输方式: stdio
分类: 搜索与 AI 记忆
你需要填写：GEMINI_API_KEY, AZURE_OPENAI_API_KEY

安装

选择你使用的 AI 应用，然后复制设置。

安装之前

基础检查：1 项中通过 1 项
在你的电脑上运行，拥有你的用户账户的权限（文件、网络、应用）。
需要你自己填写： GEMINI_API_KEY, AZURE_OPENAI_API_KEY

标为机密的值是你账户的钥匙。千万不要分享给别人，也不要粘贴到聊天里。

你需要

PyPI已安装 Python 的 uv（用于 uvx 命令）

你用的是哪个应用？

ClaudeClaude CodeCursorVS CodeChatGPT

在上方选择你的应用。我们会记住它，下次按钮会显示“安装到 …”。

打开 Claude Desktop → Settings → Developer → Edit Config，把下面的内容粘贴到 claude_desktop_config.json

{
  "mcpServers": {
    "ep": {
      "command": "uvx",
      "args": [
        "ep-mcp",
        "serve",
        "--config",
        "your-value",
        "--transport",
        "stdio"
      ],
      "env": {
        "GEMINI_API_KEY": "your-value",
        "AZURE_OPENAI_API_KEY": "your-value"
      }
    }
  }
}

在终端中运行这条命令

claude mcp add ep --env GEMINI_API_KEY=your-value --env AZURE_OPENAI_API_KEY=your-value -- uvx ep-mcp serve --config your-value --transport stdio

运行前请先填写空着的值。

这个服务器需要在启动命令中填入值，请复制设置后自行填写。

或者自己复制设置

粘贴到项目中的 .cursor/mcp.json，或 ~/.cursor/mcp.json

{
  "mcpServers": {
    "ep": {
      "command": "uvx",
      "args": [
        "ep-mcp",
        "serve",
        "--config",
        "your-value",
        "--transport",
        "stdio"
      ],
      "env": {
        "GEMINI_API_KEY": "your-value",
        "AZURE_OPENAI_API_KEY": "your-value"
      }
    }
  }
}

浏览器会先询问是否打开应用，然后应用会显示设置并请你确认。密钥的值会留空，由你自己填写。

这个服务器需要在启动命令中填入值，请复制设置后自行填写。

或者自己复制设置

粘贴到项目中的 .vscode/mcp.json

{
  "servers": {
    "ep": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "ep-mcp",
        "serve",
        "--config",
        "your-value",
        "--transport",
        "stdio"
      ],
      "env": {
        "GEMINI_API_KEY": "your-value",
        "AZURE_OPENAI_API_KEY": "your-value"
      }
    }
  }
}

浏览器会先询问是否打开应用，然后应用会显示设置并请你确认。密钥的值会留空，由你自己填写。

这个服务器在你的电脑上运行，而 ChatGPT 只能连接互联网上的服务器。请选择其他应用。

你的选择通过 Cookie 保存在这台设备上，无需账户。

需要你填写的值

GEMINI_API_KEY 机密 env
Gemini API key for embeddings (or set AZURE_OPENAI_API_KEY when using Azure OpenAI)
AZURE_OPENAI_API_KEY 机密 env
Azure OpenAI key, if the config uses the Azure embedding provider

这些设置是根据作者在注册中心的条目生成的。如果 README 的写法不同，请以 README 为准。

它能做什么

它在你的电脑上运行；安装后，工具列表会显示在你的 AI 应用中。

基础检查i

我们检查的是任何人都能看到的信息；我们没有阅读代码。不适用的检查项不计入总数。 · 检查于 2026年11月7日

✓ 通过
作者与仓库一致
注册名称已通过 GitHub 账户 brianhearn 验证，该账户拥有此仓库
… 尚未检查
仍在维护
我们还没有从 GitHub 读取这个仓库
… 尚未检查
有许可证
我们还没有从 GitHub 读取这个仓库
– 不适用
端点有响应
它在你的电脑上运行，没有互联网端点
– 不适用
使用 OAuth 登录
在你的电脑上运行，改用本地设置
… 尚未检查
工具描述中没有隐藏指令
它在你的电脑上运行；我们只能读取互联网端点的工具列表

与“搜索与 AI 记忆”中的其他服务器比较i

星标数多于此分类中其他 7270 个服务器的 0%
发布到注册中心的时间比此分类中其他 7270 个服务器的 69% 更新
基础检查结果等于或优于此分类中其他 7270 个服务器的 100%

与注册中心里这个分类中所有活跃的服务器比较。只有当发布者拥有该仓库时，星标数才计入。

常见问题

ExpertPack MCP 安全吗？

它在 1 项适用的基础检查中通过了 1 项。我们没有阅读代码。在添加密钥或令牌之前，请先看看这个服务器要求哪些权限，并使用只具备所需权限的密钥。

它需要哪些密钥或代码？

你需要填写：GEMINI_API_KEY, AZURE_OPENAI_API_KEY

它在我的电脑上运行还是在云端？

它在你的电脑上运行。你的 AI 应用会在需要时启动它，数据不会经过作者的服务器（除非它自己调用在线服务）。

它还在维护吗？

GitHub 仓库最后更新于–，最新版本（0.7.0）于17天前进入注册中心。

哪些应用可以使用它？

支持 MCP 的应用，例如 Claude Desktop、Claude Code、Cursor 和 VS Code。你的电脑上还需要有 Python (uv)。

有什么可以替代它？

同一分类中有：Context7, Codebase Memory, Exa

替代选择

同一分类中其他作者的服务器，按星标数排序。

如何看卡片：★ GitHub 星标i三大支柱：评价 · 检查 · 使用情况i更新 = 最新版本发布到注册中心的时间i

Context7

作者：upstash

Up-to-date docs for AI agents

电脑或云端★ 64,224MIT昨天更新

安装

Codebase Memory

作者：DeusData

Codebase knowledge graph for AI agents — 162 languages, sub-ms queries, 99% fewer tokens.

在你的电脑上运行★ 48,148MIT21天前更新

安装

Exa

作者：exa-labs

Connect AI agents to Exa for web search, content fetching, and multi-step research.

在云端运行★ 5,213MIT上个月更新

安装

作者的 README

打开 README

按原文从作者的仓库复制。PopNx MCP 没有审查或背书其内容。HTML、图片和脚本已被移除。链接会离开本站。

ep-mcp

Expertise-as-a-Service over the Model Context Protocol.

An MCP server that turns any ExpertPack into a live, queryable knowledge service. Any MCP-compatible agent (Claude Desktop, Cursor, Windsurf, Claude.ai, custom hosts) can connect and retrieve high-quality, provenance-rich domain expertise via standardized tools.

[🌐 expertpack.ai](https://expertpack.ai) · [📦 ExpertPack Framework](https://github.com/brianhearn/expert-pack) · [📖 Architecture](ARCHITECTURE.md)

97.6% retrieval hit rate / 0 full misses on a 22-question benchmark against a production v4.1 atomic-conceptual pack (658 chunks). Prior v4.0 baseline on the same pack and eval set: 65.1%.

Features

Schema-aware hybrid retrieval: BM25 + vector search (sqlite-vec), metadata boosting, MMR re-ranking, adaptive threshold filtering, length penalty
`requires:` expansion (schema v4.1 atomic-conceptual): When a top-K atom's frontmatter declares requires: [B, C], the referenced atoms are auto-appended to results as additional SearchResults flagged requires_expanded=True. Directional (A→B does not imply B→A), transitive up to 2 hops, capped at 3 atoms and 3,500 cumulative tokens, appended after the max_results slice so it never displaces top-K.
Intent-aware routing: Automatic query classification (ENTITY/HOW/WHY/WHEN/GENERAL) adjusts vector/BM25 fusion weights per query — no LLM required
Query embedding cache: SQLite-backed cache keyed by (model, dimension, sha256(query)) — warm query latency drops from ~4s (live Gemini API) to ~1ms
Deep graph traversal: Optional multi-hop BFS expansion over knowledge graph edges with per-hop score decay and drift prevention
Reserved-slot graph merge: Configurable protected slots in the final top-K reserved for graph-expanded neighbors so related files surface even against dominant core results
BM25 fallback mode: Embedding provider outage degrades gracefully to lexical-only search (coverage + density + path + length scoring) instead of a hard error
Cross-encoder reranker: Optional second-pass precision reranker (disabled by default, ~70ms warm latency when enabled)
Provenance-first: Every result includes id, content_hash, verified_at, source_file
Reconstruct mode (RFC-003): Opt-in reconstruct=true / reconstruct: true enriches results with fragment_id, line_range, original_markdown, span-level content_hash, excerpt, stale, and byte offsets so agents can prove exactly what source text grounded an answer
EP-native chunking: Files are treated as atomic retrieval units (split at ##/### only for oversized content). Split chunks keep the original span as content/span_hash, carry a computed line_range, and record a heading_path (0.7.0+)
Heading paths (0.7.0+): every hit carries heading_path (e.g. ["Routing", "Setup", "Ports"]) so an H3 hit keeps its H2 parent; ep_read(path, section=..., neighbors=true) reads just that section plus its previous/next siblings
Freshness (0.7.0+, ExpertPack Core 4.2): hits and ep_read carry verified_age_days, age_stale (verified_at older than the file's refresh or the manifest freshness.refresh_cycle), expired and expires_at (volatile TTL). stale keeps its old meaning — index drift (file changed since indexing, reconstruct mode only). Optional down-rank via retrieval.freshness_penalty (default 0.0 = off)
Skills over MCP (0.7.0+, opt-in): set mcp.skills.enabled: true in a pack's manifest.yaml to declare the io.modelcontextprotocol/skills extension (SEP-2640). Each pack becomes one skill at skill://{slug}/SKILL.md (generated) with every non-on_demand file served verbatim at skill://{slug}/{path}; skills/list / skills/get publish the manifest with SHA-256 digests, sizes, ttlMs and cacheScope. Host support is still thin, so the ep_* tools remain the primary path
Frontmatter aware: Strips metadata for embedding, extracts type, tags, requires, etc. for boosting/expansion/filtering
Incremental indexing: Content-hash based staleness detection — only re-embeds changed files
Multi-pack support: Path-based routing at /packs/{slug}/mcp
Graph-aware retrieval: optional post-retrieval expansion via _graph.yaml adjacency, including accepted ontology/entity nodes when graph export provides them
`context_prefix` (RFC-004): sidecar situating strings are prepended at index time for embed + BM25 only; returned/reconstruct text stays the original atom body
Tier / navigation filter: retrieval_strategy: navigation, _index.md, source-coverage.md, and context.on_demand files are excluded from the RAG pool (still readable via ep_read / resources)
Tools:
ep_search — primary hybrid retrieval tool
ep_read — load a whole atom by path or provenance id (consume-loop step 2)
ep_list_topics — browse pack structure and available content types
ep_graph_traverse
Multi-pack stdio support: All tools accept a pack parameter for routing when multiple packs are configured. Single-pack configs make pack optional (defaults to the only configured pack).
Resources:
ep://{slug}/manifest
ep://{slug}/authority
ep://{slug}/files
ep://{slug}/file/{+path} (raw content with frontmatter)
skill://{slug}/SKILL.md, skill://{slug}/{+path} (only when mcp.skills.enabled: true)
Transports: Streamable HTTP (primary, cloud-ready), stdio (local dev)
stdio multi-pack support (0.6.0+): When using --transport stdio with multiple packs configured, all tools accept an optional pack parameter (slug string) to specify which pack to query. For single-pack configurations, the pack parameter is optional and defaults to the only configured pack. This allows MCP hosts (Cursor, Claude Desktop) to use a single stdio process for multiple packs instead of requiring one process per pack.
HTTP endpoint: GET /search (for non-MCP HTTP clients)
Golden retrieval eval (0.7.0+): tests/eval/ — YAML query sets (expected best-first, must_not, category) scored as recall@1/3/5, MRR and must_not violations per pack and category. Skipped in the default run; pytest tests/eval -m golden --golden --packs-root with EP_EVAL_TODAY=2026-10-20 and optional EP_EVAL_MIN="recall@5=0.9,mrr=0.7", or python tests/eval/golden_eval.py --golden --today 2026-10-20
Auth: API key (Phase 1), designed for OAuth 2.1 (Phase 2)

Quick Start

Installation

git clone https://github.com/brianhearn/ep-mcp.git

cd ep-mcp pip install -e .[dev]

Configuration

Create config.yaml with at minimum:

server:

host: "127.0.0.1" port: 8000 log_level: "info"

Required for public Host headers (SDK v2 DNS-rebinding protection).

See DEPLOYMENT.md — omit and Streamable HTTP returns 421.

mcp_allowed_hosts:

"localhost"
"localhost:*"
"127.0.0.1"
"127.0.0.1:*"

packs:

slug: "my-pack" path: "/path/to/your/expertpack" api_keys:
"your-secret-key-here"

embedding: provider: "gemini" # "gemini" (default) or "azure-openai" model: "gemini-embedding-001"

For the full configuration reference — embedding providers, retrieval tuning (mmr_lambda, min_score, default_max_results, graph expansion, reranker, requires: expansion, etc.), production systemd setup, and deployment gotchas — see [DEPLOYMENT.md](DEPLOYMENT.md).

Running

ep-mcp serve --config config.yaml

The server builds/updates the SQLite index (FTS5 + sqlite-vec) on startup. 0.5.0 requires a rebuild of existing indexes (index_features=context_prefix_v1+nav_filter_v1); startup does this automatically, or delete /.ep-mcp/ first.

The GET /search HTTP endpoint is available at:

GET /search?q= &pack= &n=10

Add &reconstruct=true to include original markdown spans and provenance blocks for verification.

with header Authorization: Bearer .

Environment Variables

Only the key for your configured embedding provider is required.

Variable	Required	Description
`GEMINI_API_KEY`	If `embedding.provider: gemini`	Gemini embedding API key
`AZURE_OPENAI_ENDPOINT`	If `embedding.provider: azure-openai`	Azure OpenAI resource endpoint
`AZURE_OPENAI_API_KEY`	If `embedding.provider: azure-openai`	Azure OpenAI API key
`EP_MCP_KEY_{SLUG}`	Optional	Per-pack API key override (uppercase slug, hyphens→underscores)
`EP_MCP_REMOTE_HOST`	Optional	Deploy target for `scripts/deploy.sh`
`EP_MCP_REMOTE_SRC`	Optional	Remote source path for `deploy.sh`
`EP_MCP_REMOTE_SITE_PKG`	Optional	Remote site-packages path for `deploy.sh`

Tech Stack

Language: Python 3.12+
MCP Framework: official mcp SDK v2 (MCPServer, spec 2026-07-28, dual-era with 2025 clients)
Storage: SQLite with FTS5 + sqlite-vec
Embeddings: Gemini (default), configurable Azure OpenAI / OpenAI providers
Web: Starlette + Uvicorn (for Streamable HTTP)
CLI: Click

See pyproject.toml for exact dependencies.

Project Structure

See ARCHITECTURE.md for detailed module breakdown, retrieval pipeline, indexing strategy, and design decisions.

[expertpack.ai](https://expertpack.ai) — Framework website, community packs, and documentation
[ExpertPack](https://github.com/brianhearn/expert-pack) — Schema framework, validation tools, Obsidian compatibility, community packs
[ExpertPacks](https://github.com/brianhearn/expert-packs) — Published knowledge packs (private)
EasyTerritory MCP (future) — Domain-specific layer built on top of this repo + ezt-designer pack

License

Author

Brian Hearn — Built in 3 days (April 10–11, 2026).