# MCP Integration Overview - [What is MCP](#what-is-mcp) - [Why MCP in artisan](#why-mcp-in-artisan) - [10 Substrate Tools](#substrate-tools) - [Plugin-Contributed Tools](#plugin-contributed-tools) - [State File Contract](#state-file-contract) - [Architecture Diagram](#architecture-diagram) - [Related](#related) --- ## What is MCP The [Model Context Protocol](https://modelcontextprotocol.io) (MCP) is an open standard that defines how AI clients (Claude Code, Cursor, Windsurf, and similar) discover and invoke tools exposed by a local or remote server over a JSON-RPC transport. A client connects once, calls `initialize` to receive the full tool catalog, and then invokes individual tools by name. The server handles routing, argument validation, and error formatting, returning structured content the client model can reason over. MCP decouples the tool surface (what the agent can do) from the tool implementation (how it is done), so the same server can serve multiple AI clients without code duplication. --- ## Why MCP in artisan LLM agents working on Flutter codebases traditionally had one interaction mode: read source files and shell out to CLI commands. That works for static analysis and compilation, but it breaks down the moment the agent needs to observe or control a *running* Flutter application. Capturing a Semantics snapshot, tailing HTTP traffic, evaluating an expression in the live isolate, or hot-reloading after a code change all require a live connection to the Dart VM Service. Passing raw VM Service WebSocket URIs and protocol bytes through shell commands is fragile and opaque. artisan's MCP integration solves this by surfacing the running Flutter app's capabilities as first-class MCP tools. The agent calls `artisan_start` to launch the app, `artisan_reload` to push a hot reload, and plugin-contributed tools to snap the widget tree or tail HTTP requests, all without copy-pasting commands or parsing raw output. The MCP server handles VM Service discovery, lazy reconnect across stop-restart cycles, and structured error formatting so the model can self-correct. --- ## 10 Substrate Tools The MCP server always registers the following ten tools, derived from the artisan CLI's built-in command set. Tool names are normalized from `cmd:name` to `cmd_name` so they satisfy the MCP identifier constraint (`[a-zA-Z][a-zA-Z0-9_]*`). These tools run in-process via the artisan registry; nine of them require no VM Service connection (so `artisan_start` works even before any Flutter app is running), while `artisan_tinker` dispatches over the VM Service and lazy-reconnects on first call. | Tool | Maps to CLI command | Purpose | |---|---|---| | `artisan_start` | `start` | Launch the Flutter app via `flutter run` and write `state.json` | | `artisan_stop` | `stop` | Send SIGTERM to the running Flutter process and delete `state.json` | | `artisan_status` | `status` | Read `state.json` and return the current process metadata as JSON | | `artisan_logs` | `logs` | Stream the most recent stdout lines captured from `flutter run` | | `artisan_restart` | `restart` | Full restart (equivalent to `R` in the flutter run TTY) | | `artisan_reload` | `reload` | Hot reload the running app without losing widget state | | `artisan_hot_restart` | `hot-restart` | Hot restart the running app, resetting ephemeral state | | `artisan_doctor` | `doctor` | Run `flutter doctor` and return the diagnostics report | | `artisan_list` | `list` | List all registered artisan commands with their signatures | | `artisan_tinker` | `tinker` | Evaluate Dart expression in running app via VM Service. Maps to: tinker command. Requires running app (artisan_start first). | Commands intentionally excluded from the MCP allowlist: interactive commands (`help`), codegen commands (`make:*`, `*:refresh`), installer commands (`plugin:*`, `install`), and MCP meta commands (`mcp:*`). They either require a TTY, mutate source files better handled by the client's own file tools, or recurse into the MCP server itself. --- ## Plugin-Contributed Tools Beyond the ten substrate tools, plugins extend the MCP catalog by overriding `ArtisanServiceProvider.mcpTools()`. The default implementation returns an empty list; plugins return a list of `McpToolDescriptor` instances that the MCP server collects at initialize time and registers alongside the substrate tools. The same `McpFilterConfig` allow/deny rules apply uniformly, so a deny rule against a plugin tool name works identically to `tools.deny: [artisan_start]`. Plugin tools dispatch through the VM Service (not in-process), so they require a running Flutter app. The MCP server lazy-reconnects on each dispatch call, meaning the agent can call `artisan_start` first and the very next plugin tool call picks up the new app automatically. **Plugin registration is explicit.** The consumer's `bin/dispatcher.dart` wrapper must list each provider in the `baseProviders` list it passes to `runArtisan(...)` (auto-discovery via `lib/app/_plugins.g.dart` handles this for `plugin:install`-registered packages). Consumers invoke `./bin/fsa mcp:serve` (post-install via `mcp:install`) or `dart run :dispatcher mcp:serve` (Windows / no-fsa fallback) to start the MCP server with plugins collected. The legacy `bin/mcp.dart` entry point is preserved for backward compatibility and loads only the substrate commands without plugin tools. The two sibling packages that ship production plugin tools are: | Package | MCP tool reference | |---|---| | `fluttersdk_dusk` | [fluttersdk.com/dusk/mcp/tool-reference](https://fluttersdk.com/dusk/mcp/tool-reference) | | `fluttersdk_telescope` | [fluttersdk.com/telescope/mcp/tool-reference](https://fluttersdk.com/telescope/mcp/tool-reference) | Each package's provider is wired by the consumer (`DuskArtisanProvider`, `TelescopeArtisanProvider`) registered in the wrapper's `artisanProviders` list. See the per-package CLAUDE.md for registration details and each plugin's tool reference site for the current tool catalog. `magic_tinker` remains the CLI REPL host for `dart run artisan tinker`, but its MCP surface is now the substrate's `artisan_tinker`. --- ## State File Contract `artisan start` writes a single JSON file at `~/.artisan/state.json` (the path is machine-local and user-scoped). The MCP server reads this file during `initialize` to discover the running Flutter app's VM Service WebSocket URI. If the file is absent or lacks a `vmServiceUri` key, the server stays online and registers all tools, but VM Service dispatch calls return an actionable error until the app is started. The state file is a single-slot store: starting a second app overwrites the previous record. It is never committed to version control (`.gitignore` / `.pubignore` exclude `.artisan/`). Fields (from `lib/src/state/state_file.dart`): | Field | Type | Notes | |---|---|---| | `pid` | `int` | PID of the `flutter run` process | | `vmServiceUri` | `string` | Canonical `ws://host:port//ws` used for VM Service connection | | `webPort` | `int` | `--web-port` value passed to `flutter run` | | `vmServicePort` | `int` | Informational; defaults to 8181 | | `startedAt` | `string` | ISO 8601 UTC timestamp | | `profile` | `string` | `debug` or `static` | | `projectRoot` | `string` | Absolute path to the consumer project | | `device` | `string` | `chrome`, `macos`, `linux`, `windows`, or a device UDID | | `chromePid` | `int` or `null` | Chrome process PID when `--device=chrome` (D6 capture) | | `tmpProfileDir` | `string` or `null` | Temp Chrome profile directory path (D6 capture) | --- ## Architecture Diagram ``` Claude Code (MCP client) | | stdio JSON-RPC (initialize / tools/call) | ./bin/fsa mcp:serve (or: dart run :dispatcher mcp:serve) | McpServer (mcp:serve) | +-- substrate tools (artisan_start, artisan_stop, ...) | | | v | ArtisanRegistry (in-process command dispatch) | +-- plugin tools (dusk_*, telescope_*) + artisan_tinker | | VM Service WebSocket | ws://localhost:PORT//ws | VmServiceClient | | ext.dusk.* / ext.telescope.* / VM evaluate RPC | Flutter app isolate (debug mode) | +-------+-------+ | | DuskIntegration TelescopeIntegration (widget tree) (HTTP + logs + exceptions) ``` **Flow summary:** 1. The MCP client (Claude Code) connects to the server over stdio and calls `initialize`. 2. `McpServer` reads `~/.artisan/state.json` and opens a VM Service WebSocket to the running Flutter app. 3. All registered tools become available to the client model. 4. Substrate tool calls (prefix `artisan:`) route in-process through `ArtisanRegistry` without VM Service involvement, so `artisan_start` works even when no app is running. 5. Plugin tool calls dispatch through `VmServiceClient.callServiceExtension`, reaching the Flutter app's registered extension handlers in the debug isolate. 6. Results return as `CallToolResult` text content; errors carry `isError: true` with an actionable message so the model can self-correct without a human in the loop. --- ## Related - [Setup guide](setup.md): install the MCP server, configure `.artisan/mcp.json`, and wire provider registration in `bin/artisan.dart`. - [Tool reference](tool-reference.md): per-tool input schema, example calls, and error codes for all ten substrate tools, plus links to each plugin's own MCP tool reference site.