Skip to main content

Code documentation generator

Codebase docs that write themselves from GitHub

Connect a repo. Moxie generates source-cited docs on the architecture, conventions, and flows already in your code, keeps them current on every merge, and serves the same context to AI agents over MCP.

Quick Answer: Moxie Docs generates codebase documentation automatically from a connected GitHub repository. On the first index it plans an outline and writes each page - architecture, conventions, cross-file flows, and API surface - from current source, with every claim linked to the file it came from. Each merge re-checks the affected pages for drift, gaps, and thin coverage, and the same index is served to AI coding agents over MCP.
  • From source. Architecture, conventions, and walkthroughs from the code that exists today, cited line by line.
  • Searchable. Generated pages and the Markdown already in your repo, indexed together in one workspace.
  • Agent-ready. Cursor, Claude Code, and Codex pull the same index over MCP.

What you get

What lands in the workspace

Generated from source

Architecture, conventions, and flow walkthroughs written from your default branch, every claim linked to the file it came from.

Architecture map

How services, modules, and data flow fit together, with Mermaid diagrams for the paths that span files.

Convention capture

The patterns your repo already enforces - test runner, error contract, folder layout - pulled into an AGENTS-style reference.

One searchable workspace

Generated pages and your hand-written Markdown indexed together and searchable from a single place.

Current on merge

Every merge re-checks the affected pages for drift, thin coverage, and undocumented surface.

MCP for agents

Cursor, Claude Code, and Codex pull the same conventions, docs, and open gaps before the first edit.

How it works

From connect to searchable docs

  1. 01

    Connect GitHub

    Install the app on the repos you want documented. Read access to code and metadata; nothing is written without a pull request.

  2. 02

    First index

    Moxie plans an outline, then writes each page in its own pass against current source. Minutes for most repos.

  3. 03

    Search and share

    Browse the workspace, search across generated and hand-written docs, and send a teammate a link.

  4. 04

    Stays current

    Every merge re-checks the touched pages. Friday Cleanup batches fixes into one docs-only PR on Pro and Team.

What gets written

The docs a new hire actually asks for

Moxie writes structural documentation - how the system is put together and why - not a generated entry for every function. These are the pages people open in their first week and keep open after.

  1. 01

    Architecture overview

    How the codebase splits into services and modules, what each one owns, and where the boundaries sit.

  2. 02

    Cross-file flows

    The request path, the job pipeline, the auth handshake - traced through every file they touch, with a diagram.

  3. 03

    Conventions reference

    The rules the code enforces in practice: naming, test commands, error handling, and where new code is expected to go.

  4. 04

    API and interface surface

    Route handlers, exported helpers, and public types, each with its signature and where it is called from.

  5. 05

    Setup and local dev

    The install and run sequence, derived from the scripts and config that exist today rather than a year-old README.

  6. 06

    Directory guide

    What lives where, so search has somewhere to land and agents stop guessing at paths.

How generation works

Outline first, then one pass per page

Generation is not a single prompt over the whole repo. Moxie builds an outline sized to the codebase, then writes each page on its own so detail never gets spread thin - and grades its own output before you see it.

  1. 01

    Index the repo

    Moxie reads the default branch: source, config, existing Markdown, and commit history.

  2. 02

    Plan an outline

    It decides which pages the codebase warrants and how they nest, scaled to the size of each subsystem.

  3. 03

    Write each page

    Every page gets its own generation pass against the real files it describes.

  4. 04

    Cite every claim

    Each section links the source path it came from, so a reader - or a reviewer - can check it.

  5. 05

    Grade its own coverage

    A thin-docs check catches any page too light for the code it covers before it reaches the workspace.

After the first index

Generation is the start, not the deliverable

A one-time doc dump is stale within a sprint. What makes generated docs worth trusting is what keeps them true afterward.

Moxie re-indexes the surface a pull request touches when it merges, compares the new code against the pages that cite it, and records anything that no longer matches. Between Cleanup PRs every finding is already visible in the workspace and posted to the relevant GitHub pull request as an advisory check.

  • Re-checked on merge. The pages a change touches are re-evaluated for drift, gaps, and thin coverage while the change is still fresh.
  • Reviewed by humans. Fixes arrive as a docs-only Friday Cleanup PR on Pro and Team. Moxie opens pull requests; your team merges.
  • Served to agents. The same index answers MCP queries from Cursor, Claude Code, and Codex, so generated context and agent context never diverge.
  • Tunable. Dismiss a finding, mute it for 7 or 30 days, or turn a detector off for the whole repo.

At a glance

Where automated generation fits

Hosted doc platforms and symbol extractors solve different halves of the problem. Moxie generates the structural layer and couples it to the merge.

Source of content

Hosted doc platforms

You write every page by hand

Symbol extractors

Signatures and doc comments

Moxie

Architecture and flows, cited to source

Keeps up with code

Hosted doc platforms

Only when someone edits it

Symbol extractors

Symbol list regenerates; prose does not

Moxie

Re-checked on every merge

Reading level

Hosted doc platforms

Whatever the author had time for

Symbol extractors

One entry per symbol

Moxie

Structural - how the system fits together

Undocumented code

Hosted doc platforms

Invisible

Symbol extractors

Listed but unexplained

Moxie

Flagged as a gap

Feeds AI agents

Hosted doc platforms

Copy and paste

Symbol extractors

No

Moxie

Same index, over MCP

Get started

Generate docs from your repo

Connect a repository and the first index shows you the architecture, conventions, and flows already in your code - plus what is undocumented. 14-day trial, no charge today.