reader3000-mcp

reader3000-mcp

Enables sharing Markdown documents for collaborative review with inline annotations and structured change requests.

Category
访问服务器

README

reader3000-mcp

An MCP server for READER.IIIK (reader3000.com) — share a Markdown document with a person, they read it in a clean, distraction-free reader and leave inline notes on exact passages, and those notes come back to you as structured change requests. You can leave notes too (as Mr. Robot, the reserved agent persona), revise the document, and delete it when you're done.

A surface for a review loop, not a place to keep anything important.

Install

reader3000-mcp is a standard stdio MCP server. Point any MCP host at npx -y reader3000-mcp — the one command every host wraps. Most hosts read it as a config-file entry:

{ "mcpServers": { "reader3000": { "command": "npx", "args": ["-y", "reader3000-mcp"] } } }

Some spell the same thing as a one-line command:

claude mcp add reader3000 --scope user -- npx -y reader3000-mcp   # Claude Code
codex mcp add reader3000 -- npx -y reader3000-mcp    # Codex

The --scope user flag installs it for every project. Without it, Claude Code adds the server to the current directory only. Codex installs for all projects by default.

Or run it from a checkout: point your host at node mcp.mjs.

Tools

Tool What it does
create Share a Markdown document → {link, codename, name}. Hand the link to a person.
read The document + its notes (each with passage, text, and author) + version.
annotate Add your own notes, as Mr. Robot. Additive — nobody else's notes are touched.
revise Replace the content (after applying notes). Conditional: a stale version is refused.
version Cheap "has anything changed?" poll.
delete Delete a document you created, by codename.
list The local ledger: every document this server created and still owns.

ref arguments accept either a codename from the ledger (amber-harbor) or a reader3000.com share link someone gave you.

What this server does with your data — the whole truth

  • Network surface: exactly six HTTPS calls, all to https://gateway.reader3000.com (/create /read /annotate /revise /version /delete). Nothing else ever leaves your machine; the code is one file (mcp.mjs) and greppable in one sitting.
  • The gateway tier, stated honestly: the gateway encrypts/decrypts in memory and stores nothing; the storage server holds only ciphertext. Your text does transit the gateway in plaintext — that is the trade this tier makes for you not having to run any crypto. It is a good-effort privacy posture, not a vault, and it is why the rule above exists: a review surface, not storage.
  • The ledger (~/.config/reader3000/ledger.ndjson; on Windows this resolves under your user profile): every document create makes is filed locally — codename, name, link, and the secret-bearing handle that authorizes revise/delete — so a document stays yours after the conversation that spawned it ends. That file is the keys: treat it like an unpassphrased SSH key. It is written mode 600 on macOS/Linux; on Windows your user profile's ACLs protect it. The handle never appears in any tool output; the conversation transcript only ever carries what the human you're sharing with would see anyway (the link, the codename).
  • A link is access: anyone holding a share link can read, annotate, and revise that one document (not delete it). Documents expire on the server 30 days after creation.

Configuration

Env var Default Meaning
GATEWAY_BASE https://gateway.reader3000.com The gateway to talk to (dev override).
READER3000_LEDGER $XDG_CONFIG_HOME/reader3000/ledger.ndjson Ledger file path.

Note: MCP hosts spawn servers with a minimal environment — set overrides in the host's server config (env block), not your shell profile.

Verifying

npm install, then npm test — from a checkout or the unpacked npm tarball, which ships the suite. It drives the real server over stdio against a stub gateway (no network, no reader3000 account needed) and locks the secret-hygiene rule: no tool output ever contains a handle.

This repository is published for use and audit; it does not take contributions.

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