MCP

Give agents the context they keep trying to rediscover.

Repo docs, conventions, gaps, and verified commands - compact MCP tools, same index your team reads.

  • Fewer tokens. One scoped lookup instead of dozens of file reads.
  • Never stale. Re-checked on every merge - agents see what code does now.
  • Read-only. Context only - never billing, settings, or merge.
Set up with one promptConnect GitHub

CLI on GitHub · Pricing

Works with Cursor, Claude Code, Codex, and Windsurf

  • Cursor
  • Claude Code
  • Codex
  • Windsurf

Task: add retry handling to the billing webhook

Without Moxie

read package.json

grep "webhook" · 38 matches

read api/billing/route.ts

read lib/billing/stripe.ts

read 11 more files…

grep "retry" · 24 matches

still inferring conventions

Every prompt, every task

With Moxieread-only

→ get_conventions()

12 rules · paths cited

→ get_doc_impact(billing/route.ts)

2 docs to update in this PR

→ search_docs("webhook retries")

top hit: billing webhook behavior

ready to write

Three scoped lookups

Works with the agents your team already uses

Works with Claude, Cursor, Gemini, GitHub Copilot, VS Code, Codex, Antigravity, Windsurf, Cline, Zed, JetBrains, Neovim, Warp.

The toolbox

Ten tools, one continuously updated index.

Everything an agent should know before editing, and the tools to land doc updates in the same PR as the code change.

  • moxie.get_conventions

    Returns categorized repo conventions, confidence, and evidence paths agents should follow before editing.

  • moxie.search_docs

    Searches generated docs and source-linked explanations for implementation details, APIs, flows, and patterns.

  • moxie.list_docs

    Lists every generated doc as a paginated table of contents, grouped by section, so agents can browse what already exists before adding more.

  • moxie.get_doc_gaps

    Lists missing or stale documentation areas with severity, summaries, and suggested doc paths.

  • moxie.get_documentation_opportunities

    Lists recommended documentation updates agents should consider while planning or finishing code changes.

  • moxie.get_documentation_patterns

    Explains where this repo keeps docs today so generated updates follow the team's existing structure.

  • moxie.get_ai_context

    Provides compact agent-ready context: setup status, commands, conventions, docs, gaps, and team notes.

  • moxie.get_doc_impact

    Given the file paths an agent plans to change, returns the matching conventions, doc gaps, and existing docs to update in the same PR, and flags net-new or undocumented paths.

  • moxie.propose_doc_update

    Resolves where a doc should live and returns the path and content for the agent to write into its own branch, so docs land in the same PR as the code change.

  • moxie.propose_doc_removal

    Validates and returns the path of a now-irrelevant Moxie-tracked doc so the agent deletes it in the same PR as the code change.

Plus slash commands & an auto-installed skill

  • /moxie-docs:document-this-changeRuns Moxie's doc-impact check on your changed files and walks the agent through landing the doc updates in the same PR.
  • /moxie-docs:fix-stale-docsSurfaces this repo's open doc gaps and drift, then guides the agent to repair them.
  • moxie-docs skillAuto-committed to your repo. Triggers when an agent edits code and points it at live Moxie context before it opens a PR - no extra setup.

Setup

Works where your agents already work.

The fastest way

$ npx moxie-docs setup

Sign in, configure your editor, and install the skill - one command.

Open source on GitHub →

Prefer manual setup?

Wire up any MCP client by hand. OAuth-capable clients can sign in without a token; use a repo token for everything else.

  1. 01

    Create a repo token

    Encrypted. Reveal, rotate, or revoke anytime.

  2. 02

    Add the endpoint

    Copy-paste configs for each client in the dashboard.

  3. 03

    Agents read first

    Conventions, docs, and gaps before the first edit.

FAQ

MCP questions, answered.

What teams ask before pointing their agents at the Moxie Docs MCP endpoint.

Is there a one-command setup?+

Yes. Run `npx moxie-docs setup` and the open-source CLI signs you in through your browser, writes your editor's MCP config, installs the moxie-docs skill and AGENTS.md block, and smoke-tests the connection. By default it writes a URL-only config so OAuth-capable editors like Claude Code, Cursor, and VS Code sign in themselves - no token to paste. Pass `--token` if you'd rather use a static repository token instead.

Is the MCP endpoint included in my plan?+

Yes. The MCP context endpoint is included on every plan - Starter, Pro, and Team. You can mint a repository-scoped token as soon as a repo finishes its first index.

What can an agent not do through MCP?+

Anything that changes state. The tools are read-only context - conventions, docs, gaps, and verified commands. An agent can't merge, edit settings, touch billing, or push code through Moxie. Doc updates it proposes still arrive as reviewable PRs you control.

How do I revoke or rotate a token?+

From your dashboard, anytime. Tokens are stored encrypted, so you can reveal, rotate, or revoke them - and any agent using a revoked token immediately stops receiving context.

Which agents and editors can connect?+

Any MCP-compatible client or editor - Claude Code, Cursor, Codex, VS Code, Windsurf, and more. Your dashboard generates copy-paste configuration for each one.

Your agents are re-reading the repo right now.

Connect a repo and the MCP endpoint goes live with the first index.

Connect GitHub repoTrial details