Moxie Docs MCP

Moxie Docs MCP

Indexes GitHub repositories and provides living documentation context to coding agents, enabling them to ground work in actual repository conventions and propose doc updates.

Category
访问服务器

README

Moxie Docs — MCP & Agent Tools

npm version

The Moxie Docs Model Context Protocol (MCP) server, the agent skills Moxie installs into your repository, and the moxie-docs CLI that wires it all up — npx moxie-docs setup.

This repo is also the open-source home of the moxie-docs npm package: a setup CLI plus the MCP interface (tool catalog, prompts, server metadata, and skill renderers) you can import. The hosted server, indexing pipeline, and your generated docs stay in the private Moxie service.

Moxie Docs indexes a connected GitHub repository and keeps its documentation alive — generating docs, discovering conventions, tracking documentation gaps and drift, and opening pull requests that keep docs in sync with code. The MCP server exposes that living context to coding agents (Claude Code, Cursor, VS Code, Codex, and any MCP-capable client) so they ground their work in how your repository actually does things — and keep the docs current as they change code.

  • Website / dashboard: https://moxiedocs.com
  • Human docs for the MCP server: https://moxiedocs.com/mcp
  • MCP endpoint: https://moxiedocs.com/api/mcp

This repository is documentation and distribution assets only. The hosted server, indexing pipeline, and your repositories' documentation are not stored here.


Contents


Quick start

The fastest path — from your repository's directory:

npx moxie-docs setup

This signs you in (OAuth in your browser), writes the MCP config for your editor, installs the moxie-docs skill + AGENTS.md guidance, and verifies the connection. After that, your agent reads live context with moxie.get_ai_context before editing and proposes doc updates that land in the same PR.

Prefer to do it by hand? Connect a repository at https://moxiedocs.com, then add the server to your agent with a config block.


CLI

The moxie-docs CLI (Node ≥ 18, zero runtime dependencies) sets up and manages the connection. Run it with npx moxie-docs <command> or install it globally (npm i -g moxie-docs).

Command What it does
moxie-docs setup End-to-end: sign in, configure your MCP client, install the skill + AGENTS.md block, and smoke-test.
moxie-docs login Sign in via the browser (OAuth 2.1 + PKCE) and store credentials in ~/.moxie-docs/.
moxie-docs status Show your scope, connected repositories, and available tool count.
moxie-docs config <client> Write the moxie-docs server entry into a client config (claude-code, cursor, or vscode).
moxie-docs install-skill Install the moxie-docs skill and AGENTS.md guidance into the current repo.
moxie-docs logout Remove stored credentials.

Useful flags: --client <name>, --token <dashboard-token>, --repo owner/name, --no-browser, --yes.

How authentication is written. By default setup/config write a URL-only server entry and let your editor perform its own OAuth sign-in on first use — so nothing in your committed config expires. The CLI's own browser sign-in (used for status and the setup smoke-test) stores a short-lived token plus a refresh token under ~/.moxie-docs/ (file mode 0600); it is never written into a client config. If you'd rather pin a static Authorization header, pass a long-lived dashboard token with --token and the CLI writes that instead. The CLI never logs token values, and it refuses to overwrite a client config it can't safely parse rather than discarding your other MCP servers.


The MCP server

Connection

Transport Streamable HTTP MCP
Endpoint https://moxiedocs.com/api/mcp
Protocol version 2025-06-18
Repository selection Tools accept an optional repository argument (owner/name, e.g. acme/app). It is required only when a token serves multiple repositories and no single default applies. Matching is case-insensitive.

All write tools are proposals — Moxie returns the file path and content for you (the agent) to write into your own branch. Moxie never edits your repository directly through MCP, and never merges. Reads return compact, citation-backed context rather than raw code dumps.

Authentication

Bearer token (available today). Create a token in the dashboard and send it as a header:

Authorization: Bearer <MOXIE_TOKEN>

Moxie stores only the token hash (plus an encrypted copy so you can reveal it), a prefix, scope, and a last-used timestamp. You can reveal, rotate, or revoke tokens from the dashboard at any time.

OAuth 2.1 sign-in (for clients that support it). MCP clients that implement the OAuth authorization flow can connect by signing in through the browser instead of pasting a token — the client discovers the server's OAuth metadata, registers itself dynamically (RFC 7591), and obtains a token via the authorization-code + PKCE flow. Discovery documents:

  • Protected resource metadata: https://moxiedocs.com/.well-known/oauth-protected-resource
  • Authorization server metadata: https://moxiedocs.com/.well-known/oauth-authorization-server

Both authentication paths resolve to the same token model, so either works with the same tools. Connected applications can be reviewed and revoked from the dashboard.

Client configuration

Most agents accept a remote MCP server with a URL and headers. Replace <MOXIE_TOKEN> with a token from the dashboard.

Claude Code.mcp.json (or claude mcp add):

{
  "mcpServers": {
    "moxie-docs": {
      "url": "https://moxiedocs.com/api/mcp",
      "headers": {
        "Authorization": "Bearer <MOXIE_TOKEN>"
      }
    }
  }
}

Cursor.cursor/mcp.json:

{
  "mcpServers": {
    "moxie-docs": {
      "url": "https://moxiedocs.com/api/mcp",
      "headers": { "Authorization": "Bearer <MOXIE_TOKEN>" }
    }
  }
}

VS Code.vscode/mcp.json:

{
  "servers": {
    "moxie-docs": {
      "type": "http",
      "url": "https://moxiedocs.com/api/mcp",
      "headers": { "Authorization": "Bearer <MOXIE_TOKEN>" }
    }
  }
}

Keep tokens out of source control. Reference an environment variable where your client supports it, or paste the token only into local, untracked config.

Tools

All tools are read-only except the two propose_* tools, which return a proposal for the agent to apply in its own branch.

Tool Purpose
moxie.get_ai_context Compact pre-edit briefing: repo status, verified commands, top conventions, open gaps, team notes. Read this first.
moxie.get_doc_impact Given the paths you're about to change (and any you're deleting), returns the conventions, gaps, and existing docs whose evidence overlaps them — and flags net-new/undocumented surfaces.
moxie.get_conventions Discovered coding conventions, grouped by category, with confidence scores, agent guidance, and source-file citations.
moxie.search_docs Semantic + keyword search over generated docs, conventions, gaps, and AI context.
moxie.get_doc_gaps Unresolved documentation gaps with severity and the paths they concern.
moxie.get_documentation_opportunities Recommended doc work: missing docs, drift repairs, and PR templates.
moxie.get_documentation_patterns How the repository organizes and maintains its docs (where new docs belong).
moxie.list_docs Paginated, section-grouped table of contents of every generated doc.
moxie.propose_doc_update Add or update a doc as part of your current change; returns the target path + Markdown to write into your branch.
moxie.propose_doc_removal Remove a Moxie-tracked doc your change makes obsolete; returns the path to delete in your branch.

Prompts

The server also exposes MCP prompts that script the common workflows:

Prompt Purpose
document-this-change Runs the doc-impact check on your changed paths and walks you through proposing the doc updates so they land in the same PR.
fix-stale-docs Finds the repository's open documentation gaps and drift opportunities and repairs them through Moxie.

Skills

Agent Skills are small, triggered instruction files that teach an agent when and how to use a capability. Moxie publishes one skill, moxie-docs, that wires an agent into the edit-then-document workflow.

The moxie-docs skill

The canonical skill lives in this repo at skills/moxie-docs/SKILL.md and is also served from the website (with a published SHA-256 digest) at:

https://moxiedocs.com/.well-known/agent-skills/moxie-docs/SKILL.md

It tells an agent to read live context before editing (get_ai_context, get_doc_impact), follow the repository's real conventions and verified commands, and propose doc updates/removals that ship in the same PR as the code change — never inventing filler docs.

What Moxie installs into your repository

When Moxie connects a repository it opens one "Add Moxie agent guidance" PR that adds three self-healing, deterministic files (regenerated only when they drift or the repo is renamed — never churning on every index):

File Purpose
AGENTS.md A short, always-on pointer block (between <!-- moxie-docs:start --> / <!-- moxie-docs:end --> markers) telling any agent to use the MCP server and the moxie-docs skill. Human-added content outside the markers is preserved. See templates/AGENTS.md.
.claude/skills/moxie-docs/SKILL.md The repo-specific moxie-docs skill for Claude Code.
.agents/skills/moxie-docs/SKILL.md The same skill for .agents-aware clients.

The installed skill is the repo-specific variant of skills/moxie-docs/SKILL.md: identical workflow, with your owner/name baked in so multi-repo tokens target the right repository. Everything Moxie writes is deterministic and contains no volatile data — live conventions and commands always come from the MCP server, not the committed file.


Use as a library

The package also exports the open-source MCP interface — the backend-independent contract — so you can build tooling against Moxie without depending on the private server. It carries no secrets and no server logic.

import {
  moxieMcpTools,            // the tool catalog (names, descriptions, input schemas, annotations)
  moxieMcpPrompts,          // the document-this-change / fix-stale-docs prompt descriptors
  moxieMcpServerInfo,       // { name, version }
  moxieMcpProtocolVersion,  // "2025-06-18"
  moxieMcpCapabilities,
  renderSkillMarkdown,      // ({ repository }) => the moxie-docs SKILL.md for a repo
  renderAgentsGuidanceBlock,// ({ repository }) => the AGENTS.md guidance block
  moxieDocsSkillMarkdown,   // the canonical published skill, as a string
  MOXIE_MCP_ENDPOINT,       // "https://moxiedocs.com/api/mcp"
  MOXIE_SITE_URL,
} from "moxie-docs";

What it does not include: the tool handlers, the indexing/doc-generation pipeline, AI/vector-search, the database layer, or token storage — those remain in the hosted Moxie service.


Discovery endpoints

Moxie publishes a small machine-discoverable surface so agents and crawlers can find the server:

Endpoint What it serves
/.well-known/mcp/server-card.json MCP server card: serverInfo, protocol version, transport, capabilities.
/.well-known/api-catalog RFC 9727 catalog linking to the server card and human docs.
/.well-known/agent-skills/index.json Agent Skills discovery index (skills + SHA-256 digests).
/.well-known/agent-skills/moxie-docs/SKILL.md The published moxie-docs skill.
/.well-known/oauth-protected-resource OAuth protected-resource metadata (RFC 9728).
/.well-known/oauth-authorization-server OAuth authorization-server metadata (RFC 8414).

License

MIT © Jackalope Dev. The contents of this repository (documentation, skill files, and templates) are open source. The hosted Moxie Docs service, indexing pipeline, and generated documentation are not part of this repository.

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