Context Engine B3723191

MCP搜索检索

提供方 TeddyJubu

搜索检索 stdio · 本地进程 Token 社区 / 第三方 Claude DesktopClaudeCursorWindsurf
查看接入方式

一个纯本地的语义搜索系统,通过Chrome扩展捕获网页文档,并通过MCP服务器使VS Code编码代理可搜索。

一句话选型结论

按资料字段整理

Context Engine B3723191 属于搜索检索类 MCP 服务器,stdio · 本地进程,Token,可直接接入 Claude Desktop、Claude、Cursor 等,有可直接执行的安装命令。接入前先确认权限范围;涉及写入、删除或付款的操作,建议先在测试环境验证。

分类搜索检索 传输方式stdio · 本地进程 鉴权Token 部署方式本地部署
GitHub 星标 0 仓库热度,无公开仓库时显示未标注
工具数 3 暴露给模型的工具数量
资源 / 提示词 0 / 0 资源与提示词模板数量
维护方 社区 / 第三方 核验等级 仅资料核对

工具与能力

搜索检索
数量来自来源目录的标注,具体工具清单以官方文档为准
工具数量3暴露给模型可调用的工具个数
资源数量0可读取的资源条目
提示词数量0内置提示词模板数量

能力标签

支持客户端

适用场景

语义搜索Python本地部署本地处理文档索引开发工具集成Chrome扩展

安装与配置

stdio · 本地进程 Token
客户端在本机拉起子进程,数据不出本机
python3 connect.py

这条命令只是入口,具体启动参数请按官方文档补齐,不要直接照搬到别的服务器。

  1. 1
    准备运行环境

    stdio 服务在本地起进程,需要先装好 Python,再按安装命令拉取依赖。

  2. 2
    安装服务

    复制上面的安装命令执行;命令里带占位符的位置,换成自己的密钥或路径。

  3. 3
    写入客户端配置

    在客户端的 MCP 配置里填好启动命令和参数,保存后完全退出客户端再打开,让新配置生效。

  4. 4
    验证是否可用

    连接成功后客户端会列出约 3 个工具,随便调一个只读工具确认能返回结果。

典型使用场景

3 类
按分类与标签整理的常见用法,具体以官方文档为准
联网搜索取证

让模型查最新资料并给出出处。

资料聚合

把一个主题的网页、文档集中抓回来整理。

情报与竞品检索

按关键词持续检索行业动态和竞品信息。

选择建议

选之前先看这几条
按条目字段给出的判断,不替你决定用不用

值得选的地方

  • 资料标注 3 个工具,职责相对集中,接入后不容易误操作。
  • 有现成的安装命令,不用自己找入口,照着跑就能起服务。
  • 已适配 Claude Desktop、Claude、Cursor,换客户端不用重新研究配置。

先想清楚的地方

  • stdio 方式要求本地装好运行时(如 Node.js),换机器要重新装一遍。
  • 使用前要准备好Token对应的凭据,并按最小权限配置。
  • 版本更新与行为细节以官方仓库为准。

更适合

  • 需要在搜索检索场景里让模型直接调用外部能力的人。
  • 已经在用 Claude Desktop、Claude 的用户,接入路径最短。
  • 想先快速试一下、能接受在本地或自有环境跑服务的人。

不太适合

  • 完全不想在本地装环境的人。
  • 不方便管理密钥或第三方授权的人。
  • 需要厂商级 SLA 和技术支持的生产环境。

同类对比

6 个同类
同分类里挑最接近的几条,字段以各自详情页为准
MCP 服务器工具数传输方式部署方式安装命令最后核验
Context Engine B3723191当前 3 stdio · 本地进程 本地部署 有 2026-05-17
Brave Search 433812ff 2 未标注 未标注 无 2026-05-17
Sequentialthinking 1 SSE · 服务端推送 远程部署 有 2026-05-18
Baidu Netdisk 未标注 未标注 未标注 无 2026-05-17
ragflow 未标注 stdio · 本地进程 本地部署 有 2026-04-11
mindsdb (430e18) 未标注 未标注 未标注 有 2026-05-17

服务器档案

搜索检索
字段来自项目仓库与公开索引,中文说明由本站标注
提供方TeddyJubu上游标注的维护方 TeddyJubu
分类搜索检索本站按上游分类展示
连接方式stdio · 本地进程客户端在本机拉起子进程,数据不出本机
鉴权方式Token远程服务建议用最小权限凭据
维护方社区 / 第三方核验等级 仅资料核对
包名 / 运行时未标注 · Python本地安装时用到的包标识与运行时
部署类型本地部署上游给出的部署形态
最近核验2026-05-17上游最后核验时间

详细介绍

9.5 KB
来自项目仓库与上游目录的原文,未做删改

Context Engine

A local-only semantic search system that captures web documentation from Chrome and makes it searchable by VS Code coding agents via MCP.

No cloud, no subscriptions. Everything runs on your machine.```
Chrome Extension → FastAPI Server (localhost:11811) → MCP Server → VS Code Agent


---

## Features

- **Chrome Extension** — one-click capture of any page or selection; crawl entire doc sites
- **YouTube transcripts** — optional popup tool to add the active YouTube video's captions locally through the browser session
- **Semantic search** — `BAAI/bge-base-en-v1.5` embeddings via sentence-transformers (768-dim, fully local)
- **Multi-collection store** — separate zvec index per topic (e.g. `raycast-docs`, `react-docs`)
- **Async BFS crawler** — polite, concurrent crawling of documentation sites
- **MCP server** — plug into Claude Code, Cursor, Copilot, Continue.dev — any MCP-compatible agent

---

## Architecture```
Chrome Extension (Manifest V3)
  ├── Popup: collection picker, add page, crawl site
  ├── Content Script: extract page text / selection
  ├── Context Menu: right-click → add to collection
  └── Service Worker: relay to local server
         │ HTTP (localhost:11811)
         ▼
FastAPI Server (server.py)
  ├── Multi-collection zvec store (~/.context-engine/collections/{name}/)
  ├── sentence-transformers BAAI/bge-base-en-v1.5 (local, 768-dim)
  ├── Async BFS web crawler (httpx + beautifulsoup4)
  └── REST API
         │ stdio
         ▼
MCP Server (mcp_server.py)
  ├── tool: search_docs
  ├── tool: list_collections
  └── tool: add_memory

File Structure```

context-engine/
├── server.py # FastAPI server (port 11811)
├── crawler.py # Async BFS web crawler
├── mcp_server.py # MCP stdio server for VS Code / Claude Code
├── requirements.txt
├── install.sh # One-shot setup script
└── extension/ # Chrome Extension (Manifest V3)
├── manifest.json
├── popup.html
├── popup.css
├── popup.js
├── content.js
├── background.js
└── icons/


---

## Setup

### 1. Install dependencies```bash
git clone https://github.com/TeddyJubu/context-engine.git
cd context-engine
bash install.sh

This creates a .venv/, installs all Python deps, and prints next steps.

2. Start the server```bash

.venv/bin/python3 server.py


Verify: `curl http://localhost:11811/health`

### 3. Load the Chrome extension

1. Go to `chrome://extensions`
2. Enable **Developer mode**
3. Click **Load unpacked** → select the `extension/` folder

### 4. Connect to your coding agents```bash
python3 connect.py

This auto-detects installed agents (Claude Code, Cursor, VS Code, Windsurf, Claude Desktop) and configures them. You can also connect specific agents:```bash
python3 connect.py --claude-code --cursor
python3 connect.py --all
python3 connect.py --status # check what's connected
python3 connect.py --all --remove # disconnect all


Manual configuration

If you prefer manual setup, add to your agent's MCP config:

**Claude Code** (`~/.claude.json`):```json
{
  "mcpServers": {
    "context-engine": {
      "command": "/path/to/context-engine/.venv/bin/python3",
      "args": ["/path/to/context-engine/mcp_server.py"]
    }
  }
}

Cursor (~/.cursor/mcp.json):```json
{
"mcpServers": {
"context-engine": {
"command": "/path/to/context-engine/.venv/bin/python3",
"args": ["/path/to/context-engine/mcp_server.py"]
}
}
}


**VS Code / Copilot** (`~/.vscode/mcp.json`):```json
{
  "servers": {
    "context-engine": {
      "command": "/path/to/context-engine/.venv/bin/python3",
      "args": ["/path/to/context-engine/mcp_server.py"]
    }
  }
}

Windsurf (~/.codeium/windsurf/mcp_config.json):```json
{
"mcpServers": {
"context-engine": {
"command": "/path/to/context-engine/.venv/bin/python3",
"args": ["/path/to/context-engine/mcp_server.py"]
}
}
}


---

## API Reference

The server runs at `http://localhost:11811`.

Most endpoints require the `X-Context-Token` header. The token is auto-generated on first run and stored at `~/.context-engine/token`. Retrieve it with:```bash
cat ~/.context-engine/token
Method Path Auth Body Description
GET /health No — Status, collection count, model name
GET /collections Yes — List all collections with doc counts
POST /collections Yes {name} Create a collection
DELETE /collections/{name} Yes — Delete a collection
POST /add Yes {text, collection, source?, tags?, source_type?, metadata?} Add text (auto-chunked)
POST /search Yes {query, collection?, top_k?, filter_tags?} Semantic search
POST /crawl Yes {url, collection, max_pages?, path_prefix?} Start async crawl
GET /crawl/{task_id} Yes — Poll crawl progress

When collection is omitted in /search, all collections are searched and results merged by score.

Quick examples```bash

Set TOKEN for convenience (replace with your actual token from ~/.context-engine/token)

TOKEN=$(cat ~/.context-engine/token)

Create a collection

curl -X POST localhost:11811/collections \
-H 'Content-Type: application/json' \
-H "X-Context-Token: $TOKEN" \
-d '{"name": "raycast-docs"}'

Add a snippet

curl -X POST localhost:11811/add \
-H 'Content-Type: application/json' \
-H "X-Context-Token: $TOKEN" \
-d '{"text": "Use getPreferenceValues() to read user preferences.", "collection": "raycast-docs", "source": "https://developers.raycast.com"}'

Search

curl -X POST localhost:11811/search \
-H 'Content-Type: application/json' \
-H "X-Context-Token: $TOKEN" \
-d '{"query": "how do I read user preferences", "collection": "raycast-docs"}'

Crawl an entire doc site

curl -X POST localhost:11811/crawl \
-H 'Content-Type: application/json' \
-H "X-Context-Token: $TOKEN" \
-d '{"url": "https://developers.raycast.com/basics/getting-started", "collection": "raycast-docs", "max_pages": 50}'

Poll crawl status

curl -H "X-Context-Token: $TOKEN" localhost:11811/crawl/{task_id}


---

## MCP Tools

| Tool | Parameters | Description |
|------|-----------|-------------|
| `search_docs` | `query`, `collection?`, `top_k?` | Semantic search across indexed docs |
| `list_collections` | — | Show collections and doc counts |
| `add_memory` | `text`, `collection`, `source?` | Store a fact or note |

### Usage in an agent```
search_docs("how to create a Raycast command", collection="raycast-docs")
list_collections()
add_memory("Always use useNavigation for stack-based navigation", collection="raycast-docs")

Chrome Extension

The popup gives you full control:

  • Status dot — green when the server is reachable, red when offline
  • Collection picker — switch between collections or create new ones
  • Add this page — extracts and indexes the current page's main content
  • Add selection — indexes only the text you've highlighted
  • YouTube Transcript — extracts captions from the active YouTube video tab and stores them as youtube_transcript context
  • Crawl site — BFS crawl from the current URL up to N pages, with live progress

Right-click context menus:
- Add selection to Context Engine — on any selected text
- Add page to Context Engine — on any page

YouTube transcript notes

  • Open a supported YouTube video tab, choose a collection, then click Add transcript in the popup.
  • Transcript extraction stays local to the extension and uses the browser's live YouTube session.
  • Successful items are stored with source_type: youtube_transcript and metadata including video id, title, URL, language, generated/manual caption status, and extraction method.

Crawler Details

  • Uses httpx.AsyncClient with a concurrency semaphore of 5
  • Polite 0.5s delay between requests
  • Extracts text from → ` ` → (priority order)
  • Strips nav, footer, header, aside, script, style, svg
  • Follows only same-domain links within the specified path prefix
  • Default max_pages: 200
  • Each page is chunked (~400 chars), embedded, and deduplicated before insert

Data Storage

Collections are stored at ~/.context-engine/collections/{name}/ as zvec indexes.

Each document stores: hash, text, source, tags, timestamp, and a 768-dim float32 embedding.

Deduplication is done by SHA-1 hash of the text — identical chunks are skipped on re-add.


Requirements

  • Python 3.10+
  • ~500MB disk for the embedding model (downloaded on first run)
  • Chrome / Chromium for the extensionfastapi>=0.110 uvicorn>=0.27 httpx>=0.27 beautifulsoup4>=4.12 sentence-transformers>=2.5 zvec mcp>=1.0

Environment Variables

Variable Default Description
CONTEXT_ENGINE_DIR ~/.context-engine Data directory
EMBED_MODEL BAAI/bge-base-en-v1.5 Sentence-transformers model
CONTEXT_ENGINE_EMBED_DIM 768 Embedding dimension (must match model)
CONTEXT_TOP_K 8 Default search result count
CONTEXT_ENGINE_CHUNK_SIZE 2048 Max chunk size in characters
CONTEXT_ENGINE_CHUNK_OVERLAP 200 Overlap between chunks in characters (clamped to CHUNK_SIZE - 1)
CONTEXT_ENGINE_TOKEN (auto-generated) Auth token for write endpoints (overrides ~/.context-engine/token)

常见问题

8 问
答案由本站依据条目字段整理,不引用上游原文
Context Engine B3723191 怎么安装?

在客户端按下面的安装命令接入:python3 connect.py。装完在客户端配置里引用同一个包或端点。

Context Engine B3723191 怎么在 Claude Desktop、Claude、Cursor 里配置?

它的传输方式是stdio · 本地进程,在客户端的 MCP 配置里填好启动命令和参数,保存后完全退出客户端再打开,配置才会生效。

Context Engine B3723191 收费吗?

本站不代收费用,也没有标注具体价格。是否收费取决于服务本身和背后的接口额度,正式使用前按官方说明确认。

Context Engine B3723191 需要 API Key 吗?

资料标注需要Token认证,使用前准备好对应凭据,并按最小权限配置。

Context Engine B3723191 支持哪些客户端?

资料标注支持 Claude Desktop、Claude、Cursor、Windsurf、VS Code;其他客户端只要能配置 MCP Server,也能用同样的命令或地址接入。

Context Engine B3723191 能提供哪些能力?

资料显示约 3 个工具;能力标签包括 搜索、本地处理、Python、Claude、开发工具集成。具体工具清单以官方文档为准。

Context Engine B3723191 需要什么运行环境?

运行时为 Python;属于本地进程,需要在本机装好运行环境。

Context Engine B3723191 是本地还是远程?

它是本地进程(stdio),由客户端在本机拉起。传输方式为stdio · 本地进程。

热门专题

按标签挑服务器
全部专题 →

服务器信息以作者仓库与官方文档为准;分类、传输方式与鉴权方式的中文说明为中文整理,最近核验 2026-05-17。