search-reader-mcp
MCP server extending Jina Reader to provide web page/PDF to Markdown reading and Bocha AI/web search via a single HTTP endpoint, exposing read, search, and MCP tools.
README
search-reader-mcp
扩展 Jina Reader 镜像的项目:在 Jina Reader 基础上添加自定义搜索(bocha)与 MCP 服务,以单端口整合 HTTP 服务器对外提供 read / search / mcp / sse 能力。
- 单服务、单端口:一个进程承载全部能力(默认 18081),
docker-compose仅用于便捷启动 - 复用镜像环境:进程内复用 jina 镜像的 Chrome 抓取、依赖与运行环境,不另起炉灶
- MCP 双传输:streamable HTTP 与 legacy SSE 兼容新旧客户端
特性
| 能力 | 端点 | 说明 |
|---|---|---|
read |
GET /read/<url>(/r/<url>) |
网页/PDF → Markdown,进程内复用 jina 抓取 |
search |
GET /s/<query>(/search/ai/<query>) |
bocha AI 语义搜索(总结+参考源+模态卡+追问) |
search |
GET /search/web/<query> |
bocha 网页搜索(长摘要列表) |
mcp |
POST /mcp |
MCP 服务(streamable HTTP),工具 search/read |
sse |
GET /sse + POST /messages |
MCP legacy SSE 传输(兼容老客户端) |
快速开始
docker compose(推荐)
# 前提:宿主已有环境变量 BOCHA_API_KEY(搜索必需)
docker compose up -d --build
# 访问 http://localhost:18081
默认值集中在 docker-compose.yml,按需手工修改(端口、URL、路径等)。
docker run
docker build -t search-reader-mcp .
docker run -d --name srm -p 18081:18081 -e BOCHA_API_KEY search-reader-mcp
配置(环境变量)
| 变量 | 默认值 | 说明 |
|---|---|---|
PORT |
18081 |
监听端口(容器内外一致) |
HOST |
0.0.0.0 |
监听地址 |
BOCHA_API_KEY |
— | 必填(搜索);开发/compose 共用标准环境变量 |
BOCHA_URL |
https://api.bochaai.com |
bocha API base-url |
JINA_APP |
/app |
jina 镜像应用根目录 |
SEARCH_READER_MCP_DATA |
/app/extension/data |
持久化数据目录(compose 外挂宿主 ~/.search_reader_mcp/) |
SQLITE_PATH |
<dataDir>/cache.db |
sqlite 缓存库路径 |
LOG_DIR |
<dataDir>/.log |
日志目录(按天滚动) |
持久化:sqlite 库与 .log/ 日志外挂宿主 ~/.search_reader_mcp/,容器重建后数据仍在。
API
read(URL → Markdown)
curl "http://localhost:18081/r/https://example.com"
# 或 /read/<url>
返回目标页面的 Markdown 正文(文本/plain)。
search(bocha)
GET(路径即 query,支持 query string 高级参数):
# ai 语义搜索(快捷方式 /s,或 /search/ai)
curl "http://localhost:18081/s/今天天气如何?count=5"
curl "http://localhost:18081/search/ai/今天天气如何?count=5&freshness=oneDay"
# web 网页搜索
curl "http://localhost:18081/search/web/hello%20world?count=10&exclude=spam.com"
POST(标准 JSON body,GET/POST 交叉):
curl -X POST http://localhost:18081/search/ai \
-H 'Content-Type: application/json' \
-d '{"query":"今天天气如何","count":5,"freshness":"oneDay","include":"weather.com"}'
高级参数(web 与 ai 略有差异):
| 参数 | 说明 |
|---|---|
count |
条数上限,默认 20,越界自动钳制到 1..50 |
freshness |
noLimit(默认)/oneDay/oneWeek/oneMonth/oneYear 或 YYYY-MM-DD..YYYY-MM-DD,非法值回退 noLimit |
include |
限定站点,多个用 | 或 , 分隔;web/ai 均支持 |
exclude |
排除站点;仅 web 生效 |
summary / answer |
是否返回长摘要 / AI 总结(布尔,默认透传官网) |
响应(结构化 JSON):
ai:{ "summary", "webPages": [...], "modalCards": [...], "followUpQuestions": [...] }web:{ "webPages": [...] }(webPages[]含name/url/siteName/snippet/summary)
MCP(工具 search、read)
- streamable HTTP:
POST /mcp(协议:JSON-RPC + SSE 流,服务端生成Mcp-Session-Id) - legacy SSE:
GET /sse建流,POST /messages发请求
Claude Code 接入示例(~/.claude.json 或项目 .mcp.json):
{
"mcpServers": {
"search-reader-mcp": {
"type": "http",
"url": "http://localhost:18081/mcp"
}
}
}
工具说明:
| 工具 | 参数 | 说明 |
|---|---|---|
search |
type(默认 ai,可 web)、query、count、freshness、include、exclude |
格式化文本:AI 总结/模态卡/编号网页来源/追问 |
read |
url |
网页/PDF → Markdown 正文 |
health
curl http://localhost:18081/health
# {"service":"search-reader-mcp","status":"ok"}
开发
开发环境直接使用 jina 镜像(复用 Node 24 + Chrome + 依赖)。
# 开发容器:挂载工作区 + tsc --watch + node --watch 热重载(端口 18082 映射到容器 18081)
docker compose up dev
# 或手动:宿主构建 + 单测,容器内验证
npm install # 宿主开发/测试依赖(koa/supertest 等已在 devDependencies)
npm test # tsc + HTTP 契约测试
npm run build # 构建到 dist/
容器冒烟:docs/smoke-test.md(构建→启动→就绪→各端点断言),MCP 工具验证 scripts/mcp-smoke.mjs。
目录结构
src/
index.ts 入口(加载配置 → jina 桥接 → 服务器 → 监听)
server.ts 整合服务器(路由分发 + 请求/错误日志)
config.ts 环境变量配置
bocha/ bocha 能力层(客户端 + VO 类型)
mcp/ MCP 服务层(search/read 工具)
jina/ jina koaApp 桥接(复用抓取)
cache/ sqlite 缓存基础设施(先建库)
log/ 按天滚动文件日志
test/ HTTP 契约测试(单一 seam)
docs/ 术语(CONTEXT.md)、ADR、冒烟流程
scripts/ 冒烟辅助脚本
参考
- CONTEXT.md — 领域术语表
- docs/adr/ — 架构决策(单端口整合、search 独立、read 复用)
- docs/smoke-test.md — 容器冒烟流程
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.