Merge pull request #31 from mtrdesign/feature/docs-site

docs: Material for MkDocs site published to GitHub Pages

Author: cvetty

.github/workflows/docs.yml

@@ -0,0 +1,51 @@
+name: docs
+
+# Builds the Material for MkDocs site and deploys it to GitHub Pages.
+# Pages "Source" must be set to "GitHub Actions" once in repo Settings.
+on:
+  push:
+    branches: [main]
+    paths:                       # only rebuild when docs inputs change
+      - docs/**
+      - mkdocs.yml
+      - .github/workflows/docs.yml
+      - pyproject.toml
+  workflow_dispatch:             # allow manual rebuilds
+
+permissions:
+  contents: read
+  pages: write
+  id-token: write
+
+concurrency:                     # never run two deploys at once
+  group: pages
+  cancel-in-progress: false
+
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0         # full history → correct "last updated" dates
+      - name: Install Cairo/Pango for social cards
+        run: |
+          sudo apt-get update
+          sudo apt-get install -y libcairo2-dev libfreetype6-dev libffi-dev \
+            libjpeg-dev libpng-dev libz-dev libpango-1.0-0 libpangocairo-1.0-0
+      - uses: astral-sh/setup-uv@v5
+        with: { enable-cache: true }
+      - run: uv sync --group docs
+      - run: uv run mkdocs build --strict   # CI=true ⇒ social cards generate; --strict fails on broken links
+      - uses: actions/upload-pages-artifact@v3
+        with: { path: site }
+
+  deploy:
+    needs: build
+    runs-on: ubuntu-latest
+    environment:
+      name: github-pages
+      url: ${{ steps.deployment.outputs.page_url }}
+    steps:
+      - id: deployment
+        uses: actions/deploy-pages@v4

.gitignore

@@ -26,3 +26,9 @@ whygraph.toml
 
 # Local scratch space for in-progress plan markdown (not version-controlled).
 plans/
+
+# MkDocs build output (the site is built and deployed by CI, never committed).
+site/
+
+# Material social-plugin font/image cache.
+.cache/

Makefile

@@ -13,7 +13,7 @@ IMAGE ?= whygraph:dev
 # Name of the long-running container started by `make image-debug`.
 DEBUG_NAME ?= whygraph-debug
 
-.PHONY: help sync test scan db db-down inspect image image-test image-inspect image-debug image-debug-down
+.PHONY: help sync test scan docs docs-build db db-down inspect image image-test image-inspect image-debug image-debug-down
 
 help:  ## List available targets
 	@grep -hE '^[a-zA-Z_-]+:.*?## ' $(MAKEFILE_LIST) | sort | awk 'BEGIN{FS=":.*?## "}{printf "  %-10s %s\n", $$1, $$2}'
@@ -27,6 +27,12 @@ test:  ## Run the test suite
 scan:  ## Re-scan this repo so WhyGraph is tested against itself
 	uv run whygraph scan
 
+docs:  ## Serve the docs site locally with live reload (social cards skipped — no Cairo needed)
+	uv run mkdocs serve
+
+docs-build:  ## Build the static docs site into ./site (strict; cards skipped unless CI=true)
+	uv run mkdocs build --strict
+
 db:  ## Start the DBGate database viewer (http://localhost:8081)
 	@test -f docker-compose.yml || { echo "error: docker-compose.yml missing - run: cp docker-compose.example.yml docker-compose.yml"; exit 1; }
 	@test -f .whygraph/whygraph.db || echo "warning: .whygraph/whygraph.db missing - run 'make scan' first"

README.md

@@ -0,0 +1,68 @@
+# Run with Docker
+
+Don't want Python, Node, `gh`, and CodeGraph on your machine? WhyGraph ships as a self-contained
+image. Your host needs **only Docker**. Install a tiny shim, then it's the same `init` and `scan` as a
+native install.
+
+```bash
+curl -fsSL https://raw.githubusercontent.com/mtrdesign/whygraph/main/scripts/install.sh | sh
+
+cd your-repo
+whygraph init      # bootstrap the WhyGraph DB + write config
+whygraph scan      # crawl history + refresh CodeGraph + LLM descriptions
+```
+
+## How the shim works
+
+`install.sh` drops `whygraph` and `whygraph-mcp` shims on your `PATH`. Each one runs the published
+image against the current directory:
+
+```bash
+docker run --rm -v "$PWD:/workspace" -w /workspace ghcr.io/mtrdesign/whygraph whygraph "$@"
+```
+
+The container is **ephemeral per command** - no compose, no `docker exec`, nothing to start or stop.
+Each invocation is a fresh process against the repo you're standing in.
+
+- **Everything's in the image** - Python and WhyGraph, `git`, the GitHub CLI, and Node with the
+  CodeGraph CLI. CodeGraph indexes from the in-image binary, so there's no docker-in-docker.
+- **Per-project config just works.** Each command reads the current repo's own `whygraph.toml`,
+  `.whygraph/`, and `.codegraph/`.
+- **Files come back as yours.** The shim runs as your host user, so generated files aren't
+  root-owned and git sees matching ownership.
+
+## Credentials
+
+The shim passes your environment through. A GitHub token goes in `[scan].token` of the repo's
+`whygraph.toml` (gitignored), and the shim also forwards `GH_TOKEN` / `GITHUB_TOKEN` plus
+`ANTHROPIC_API_KEY` / `OPENAI_API_KEY` / `DEEPSEEK_API_KEY` from your environment.
+
+!!! warning "Never bake a token into the image"
+    Pass credentials at run time, never at build time. The repo's gitignored `whygraph.toml` is the
+    right home for a pinned token.
+
+## Wire your editor, still only Docker
+
+The MCP server is containerized too. `install.sh` drops a `whygraph-mcp` shim alongside `whygraph`, so
+there's nothing extra to install. Wire your editor from inside the repo:
+
+```bash
+whygraph init --agent claude     # writes .mcp.json (also: --agent cursor / vscode / codex)
+```
+
+The generated config launches `whygraph-mcp` by bare command name. Your editor resolves it to the
+shim, which starts a per-session container speaking MCP over stdio. It reads the repo's `.whygraph/`
+and `.codegraph/` over the same `/workspace` mount the scan writes to - so the editor and the scan
+share one source of truth on disk.
+
+## Build the image yourself
+
+Building locally instead of pulling - say, while developing:
+
+```bash
+docker build -f docker/whygraph/Dockerfile -t whygraph:latest .
+WHYGRAPH_IMAGE=whygraph:latest whygraph scan
+```
+
+`WHYGRAPH_IMAGE` overrides the image the shim runs, so you can test a local build without touching the
+install.

docs/deploy/docker.md

@@ -0,0 +1,30 @@
+# Docker & Self-Hosting
+
+WhyGraph ships as a self-contained image, so there are two ways to run it. Both use the same image;
+they differ in who's driving.
+
+<div class="grid cards" markdown>
+
+-   :material-laptop:{ .lg .middle } __As a local dev tool__
+
+    ---
+
+    Install the Docker shim, then `init` and `scan` your repos and wire your editor - no Python or
+    Node on the host. This is the default install.
+
+    [:octicons-arrow-right-24: Run with Docker](docker.md)
+
+-   :material-server-network:{ .lg .middle } __As a service__
+
+    ---
+
+    A containerized `whygraph-mcp` endpoint that real applications - not just editors - connect to
+    for git-based analysis of a target repo.
+
+    [:octicons-arrow-right-24: WhyGraph as a service](service.md)
+
+</div>
+
+Most people start with the local tool. Reach for the service model when you're building an
+application that needs the *why* behind code - a review bot, an onboarding assistant, an internal
+portal.

docs/deploy/index.md

@@ -0,0 +1,89 @@
+# WhyGraph as a service
+
+Editors aren't the only thing that can talk to WhyGraph. The MCP server is a plain stdio program, so
+any application that speaks MCP can connect to it for git-based analysis of a target repo. Think of a
+review bot, an onboarding assistant, or an internal dev portal that needs the *why* behind a chunk of
+code - not just the code.
+
+This page covers that model: a containerized `whygraph-mcp` endpoint a third-party app drives over
+MCP, reading the same on-disk data your scan produces.
+
+## The shape of it
+
+A consuming app launches the WhyGraph image, mounts the target repo at `/workspace`, and speaks MCP
+over stdio. The scan **writes** the databases; the MCP server **reads** them. One repo on disk is the
+single source of truth.
+
+```mermaid
+flowchart LR
+    app["Consuming app<br/>(bot · portal · assistant)"]
+    mcp["whygraph-mcp<br/>(container)"]
+    repo[("/workspace mount<br/>.whygraph + .codegraph")]
+
+    app -- "MCP over stdio" --> mcp
+    mcp -- "reads cached evidence + rationale" --> repo
+    scan["whygraph scan"] -- "writes" --> repo
+```
+
+The consuming app gets the full read surface over MCP:
+
+- **Evidence** - `whygraph_evidence_for` and `whygraph_area_history` for the commits, PRs, and issues
+  behind a path or symbol.
+- **Rationale** - `whygraph_rationale_brief` for a structured why-this-exists card.
+- **Resources** - `whygraph://commit/{sha}`, `whygraph://pr/{number}`, `whygraph://issue/{number}`,
+  and `whygraph://repo/overview`.
+
+See the [MCP surface reference](../reference/mcp.md) for exact signatures.
+
+## Scan first
+
+The server reads cached data - it doesn't crawl on demand. So the target repo must be scanned before
+an app connects, or the tools have nothing to return.
+
+```bash
+cd /path/to/target-repo
+whygraph scan
+```
+
+!!! warning "An unscanned repo returns nothing"
+    `whygraph_rationale_brief` raises an error when a target maps to no scanned commit, and the
+    evidence tools come back empty. Run `whygraph scan` (and keep it fresh with
+    [git hooks](../guide/scanning.md#keep-it-fresh)) before pointing an app at the server.
+
+## Launch the endpoint
+
+The MCP server runs from the same image as everything else. Mount the target repo and run
+`whygraph-mcp`:
+
+```bash
+docker run --rm -i \
+  -v "/path/to/target-repo:/workspace" -w /workspace \
+  ghcr.io/mtrdesign/whygraph whygraph-mcp
+```
+
+The `-i` flag keeps stdin open for the MCP stdio transport. Your app spawns this command and talks
+JSON-RPC to it, exactly as an editor would. The [`install.sh` shim](docker.md) wraps the same call as
+a bare `whygraph-mcp` on `PATH`.
+
+## Credentials
+
+What the app needs depends on what it asks for:
+
+- **Reading cached evidence and rationale needs no credentials.** It's all in the mounted databases.
+- **Generating a *new* rationale card needs an LLM key.** `whygraph_rationale_brief` calls the
+  configured provider on a cache miss. Supply the key through the environment
+  (`ANTHROPIC_API_KEY`, `OPENAI_API_KEY`, `DEEPSEEK_API_KEY`) or the repo's `whygraph.toml`
+  `[llm.*]` table.
+
+!!! info "Tokens never live in the image"
+    Pass credentials at run time via env or the gitignored `whygraph.toml` - never bake them into a
+    built image. This matches WhyGraph's own invariant for the Docker shim.
+
+## Current scope
+
+Today the server speaks MCP over **stdio, one session per process**. There's no long-running HTTP
+endpoint yet - each connection is its own `docker run`. A persistent server mode with an HTTP MCP
+transport is on the [roadmap](../roadmap.md), not built.
+
+For now, model your integration as "spawn a session, run the tools you need, let it exit" - the same
+lifecycle an editor uses, driven by your app instead.

docs/deploy/service.md

@@ -0,0 +1,61 @@
+# Getting Started
+
+WhyGraph is a rationale layer over [CodeGraph](https://github.com/colbymchenry/codegraph). CodeGraph
+maps what your code *is* - symbols, callers, callees. WhyGraph adds *why* it exists, drawn from the
+history around it.
+
+For each chunk of code, it collects evidence from git and GitHub - commits, blame, pull requests, the
+issues those PRs closed - and links it together. Then it exposes that evidence to your AI editor over
+MCP, plus an on-demand rationale card (purpose, why, constraints, tradeoffs, risks) that it caches.
+
+You install WhyGraph once, then wire it into each repo you want it to analyze. It speaks MCP, so any
+editor that does too can use it.
+
+## CodeGraph vs WhyGraph
+
+The two tools stay in their lanes:
+
+| Layer | Answers | Examples |
+|---|---|---|
+| **CodeGraph** | "What's connected to what?" | callers, callees, symbol resolution, type hierarchy |
+| **WhyGraph** | "Why does this exist, and when did it change?" | evidence, rationale cards, area history |
+
+Run both. WhyGraph reads CodeGraph's index to resolve a symbol to a file and line range, then layers
+its own history on top.
+
+## Prerequisites
+
+You don't need all of these - most are optional, and the phases that depend on them skip cleanly when
+they're missing.
+
+- **[uv](https://docs.astral.sh/uv/)** *or* **Docker** - uv for a native install, or Docker alone for
+  the [container install](installation.md). With Docker, you need nothing else on your host.
+- **git** - your repo history is the primary evidence source.
+- **Docker** *(native installs only)* - when no `codegraph` binary is on `PATH`, `whygraph scan` runs
+  CodeGraph inside the WhyGraph image to index the repo.
+- **[`gh` CLI](https://cli.github.com/)**, authenticated - only for GitHub repos, and only if you
+  enable the remote crawl. Without it, the GitHub phase is skipped.
+- **`claude` CLI** *or* an LLM API key - for per-commit descriptions and rationale cards. Both phases
+  skip cleanly if neither is available.
+
+Ready to install?
+
+<div class="grid cards" markdown>
+
+-   :material-download:{ .lg .middle } __Installation__
+
+    ---
+
+    Docker, PyPI, GitHub, or a local checkout - pick the path that fits.
+
+    [:octicons-arrow-right-24: Install WhyGraph](installation.md)
+
+-   :material-rocket-launch:{ .lg .middle } __Quickstart__
+
+    ---
+
+    Init, scan, wire an editor - the happy path in four commands.
+
+    [:octicons-arrow-right-24: Quickstart](quickstart.md)
+
+</div>