freshrss-mcp
Enables AI assistants to read and manage FreshRSS feeds, including listing subscriptions, browsing articles, marking read/unread, and starring.
README
freshrss-mcp
An MCP server that puts a small, well-typed toolset in front of a FreshRSS instance, so an AI assistant (Claude, etc.) can read and manage your feeds: list subscriptions, browse unread/starred articles, read full content, mark read/unread, star, add feeds, and mark whole streams read.
It talks to FreshRSS over its built-in Google Reader-compatible API
(/api/greader.php), so it works with any standard FreshRSS install — no
plugins or schema changes required.
The server speaks MCP over streamable HTTP on a single port and gates the
endpoint with a static bearer token. /healthz is open for liveness/readiness.
Tools
| Tool | Description |
|---|---|
whoami |
Authenticated user + instance URL — a quick connectivity/credentials check. |
list_feeds |
Subscribed feeds with category and unread count. |
list_categories |
Categories/folders with total unread count. |
list_articles |
Articles (newest first), filterable by unread/starred/all, by feed or category, with an optional contains substring filter and pagination. |
get_article |
Full HTML content of a single article by id. |
mark_read / mark_unread |
Toggle read state for one or more article ids. |
star / unstar |
Toggle the star/favorite for one or more article ids. |
mark_all_read |
Mark an entire stream (all, a feed, or a category) read, optionally only items older than N seconds. |
add_feed |
Subscribe to a new feed URL, optionally filed under a category. |
Article ids returned by list_articles/get_article are the values you pass
back to the mutating tools.
Configuration
All configuration is via environment variables (see .env.example):
| Variable | Required | Description |
|---|---|---|
FRESHRSS_URL |
yes | Base URL of your FreshRSS instance, without /api/... (e.g. https://rss.example.com). |
FRESHRSS_USER |
yes | FreshRSS username. |
FRESHRSS_API_PASSWORD |
yes | The per-user API password (see below) — not the web login password. |
MCP_TOKEN |
recommended | Bearer token clients must send. If unset, the MCP endpoint is unauthenticated — only acceptable for purely local use. |
PORT |
no | Listen port (default 8080). |
FRESHRSS_TIMEOUT |
no | Per-request timeout to FreshRSS, seconds (default 30). |
Enabling the FreshRSS API
- In FreshRSS, go to Settings → Authentication and enable "Allow API access" (the global toggle for the GReader/Fever APIs).
- Go to Settings → Profile and set an API password. This is a
dedicated password used only by API clients; it is separate from your login
password. Put it in
FRESHRSS_API_PASSWORD.
Running locally
cp .env.example .env # then edit .env with your instance details
pip install -r requirements.txt
set -a && . ./.env && set +a
python server.py
# MCP endpoint: http://localhost:8080/mcp
# Health check: http://localhost:8080/healthz
Or with Docker:
docker build -t freshrss-mcp .
docker run --rm -p 8080:8080 --env-file .env freshrss-mcp
Connecting a client
Add it as a streamable-HTTP MCP server. With the Claude Code CLI:
claude mcp add --transport http freshrss https://your-host/mcp \
--header "Authorization: Bearer $MCP_TOKEN"
Any MCP client that supports HTTP transports works the same way: point it at
https://your-host/mcp and send Authorization: Bearer <MCP_TOKEN>.
Security notes
- No secrets live in this repo. Credentials are supplied only at runtime via
environment variables.
.envis gitignored; only.env.example(placeholders) is committed. - Always set
MCP_TOKENfor any networked deployment — without it the endpoint is open to anyone who can reach it. - The server stores no data on disk; every call proxies to FreshRSS. The ClientLogin token is held only in memory and refreshed automatically on 401.
- Prefer giving the MCP its own FreshRSS user/API password so access can be revoked independently of your main account.
How it works
server.py is a single FastMCP
streamable-HTTP app. A small GReader client handles ClientLogin auth (token
cached in memory, auto-refreshed on 401) and the handful of GReader endpoints
the tools need. A Starlette middleware enforces the bearer token on everything
except /healthz.
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。