Conduit

Conduit

A self-hosted RSS reader MCP server that manages RSS subscriptions and fetches live feed content, deployable on AWS.

Category
访问服务器

README

Conduit

A self-hosted RSS reader exposed as an MCP service, deployed on AWS ECS Fargate. Authenticated via AWS Cognito OAuth and callable directly from Claude on iPhone, iPad, and desktop.

How it works

Conduit runs as a FastMCP HTTP server behind an ALB. Claude authenticates via OAuth (proxied through FastMCP to AWS Cognito), then calls MCP tools to manage RSS subscriptions and fetch feed content. Feed items are fetched live on each tool call — there is no background polling or caching layer.

Claude → OAuth (Cognito) → ALB → ECS Fargate (Conduit) → DynamoDB
                                                         → RSS feeds (live HTTP)

MCP Tools

Tool Description
add_feed(url, label?) Subscribe to an RSS/Atom feed. Validates the URL before storing.
remove_feed(url) Unsubscribe from a feed.
list_feeds() List all subscribed feeds with URL, label, and addedAt.
get_feed_items(url, limit=50) Fetch live items from one feed. Returns title, link, published, summary.
get_all_items(limit=200) Fetch headlines across all subscribed feeds (no summaries). Use for time-based filtering on the LLM side.
get_article_content(url) Fetch and extract full plain-text content of an article. Returns title, author, published, content (≤100k chars).

Project Structure

conduit/
├── src/conduit/
│   ├── server.py      # FastMCP app, OAuth wiring, tool registration
│   ├── feeds.py       # HTTP fetch + parse (feedparser), article extraction (trafilatura)
│   └── storage.py     # DynamoDB CRUD for feed subscriptions
├── infra/
│   ├── vpc.yaml
│   ├── storage.yaml   # DynamoDB table
│   ├── cognito.yaml   # Cognito user pool + app client
│   ├── networking.yaml # ALB, ACM cert, Route 53
│   ├── ecs.yaml       # ECS Fargate cluster, task def, IAM roles
│   └── params/
│       ├── dev.json
│       └── prod.json
├── seeds/feeds.opml   # Replace with your OPML export for local dev
├── tests/
├── docker-compose.yml
├── Dockerfile
└── Makefile

Local Development

Prerequisites: Docker, Docker Compose, AWS CLI (for production deploys)

# Replace seeds/feeds.opml with your own OPML export first (optional)
make local

This starts:

  • Conduit on http://localhost:8000 with AUTH_DISABLED=true
  • DynamoDB Local on port 8001
  • An init container that creates the conduit-feeds table

On startup, Conduit parses seeds/feeds.opml and pre-populates the feed list for the local user.

Connect Claude Desktop or any MCP client to http://localhost:8000/mcp.

Deployment

Dev

Five CloudFormation stacks, deployed in order:

make deploy-all ENV=dev

Or individually:

make deploy-vpc ENV=dev
make deploy-storage ENV=dev
make deploy-cognito ENV=dev
make deploy-cognito-secret ENV=dev   # stores client secret in SSM Parameter Store
make deploy-networking ENV=dev
make deploy-image                    # builds and pushes Docker image to ECR
make deploy-ecs ENV=dev

Prod

Same sequence, passing ENV=prod:

make deploy-all ENV=prod

Before deploying prod for the first time:

  1. Set Domain in infra/params/prod.json to your production domain (e.g. conduit.example.com).
  2. Ensure a Route 53 public hosted zone exists for the root domain in the same AWS account — make deploy-networking looks it up automatically.
  3. After deploy-cognito, run make deploy-cognito-secret ENV=prod to store the Cognito client secret in SSM Parameter Store at /conduit/prod/cognito-client-secret. The ECS task reads it from there at runtime.
  4. Create a Cognito user for each person who should have access: aws cognito-idp admin-create-user --user-pool-id <id> --username <email>.

Environment parameters live in infra/params/dev.json and infra/params/prod.json. The Domain parameter must match a Route 53 hosted zone in the same account.

To tear down:

make destroy-dev    # or destroy-prod, destroy-all

Authorization Model

Every MCP tool call is scoped to the authenticated user. Conduit extracts the sub claim from the Cognito JWT and uses it as a partition key (user#<sub>) in DynamoDB. This means:

  • A user can only list, add, or remove their own feed subscriptions — there is no cross-user visibility.
  • get_feed_items enforces subscription ownership: it rejects requests for feeds the calling user has not subscribed to.
  • get_article_content does not check subscriptions — any authenticated user can fetch article content from any URL, since the URL comes from feed items they already have access to.
  • In local dev (AUTH_DISABLED=true), all operations use the hardcoded identity local-dev-user. This mode must never be enabled in production.

Environment Variables

Variable Required Description
AUTH_DISABLED No Set to true to bypass Cognito. Never set in production.
DYNAMODB_TABLE Yes DynamoDB table name (conduit-feeds)
AWS_REGION Yes AWS region
COGNITO_USER_POOL_ID Prod Cognito User Pool ID
COGNITO_APP_CLIENT_ID Prod Cognito App Client ID
COGNITO_CLIENT_SECRET Prod Cognito App Client secret
DOMAIN Prod Service domain (e.g. conduit.example.com)

Development Commands

make lint       # ruff check src/
make format     # ruff format src/
make typecheck  # mypy src/
make test       # pytest

Tech Stack

  • Python 3.12, FastMCP, feedparser, trafilatura, boto3
  • AWS: ECS Fargate, ALB, ACM, Cognito, DynamoDB, Route 53, SSM, ECR
  • Infra: CloudFormation, Docker (multi-stage, non-root)
  • Quality: ruff, mypy (strict), pytest, gitleaks (pre-commit)

推荐服务器

Baidu Map

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。

官方
精选
JavaScript
Playwright MCP Server

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。

官方
精选
TypeScript
Magic Component Platform (MCP)

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。

官方
精选
本地
TypeScript
Audiense Insights MCP Server

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。

官方
精选
本地
TypeScript
VeyraX

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。

官方
精选
本地
graphlit-mcp-server

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。

官方
精选
TypeScript
Kagi MCP Server

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。

官方
精选
Python
e2b-mcp-server

e2b-mcp-server

使用 MCP 通过 e2b 运行代码。

官方
精选
Neon MCP Server

Neon MCP Server

用于与 Neon 管理 API 和数据库交互的 MCP 服务器

官方
精选
Exa MCP Server

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。

官方
精选