- crates/mcp (L5): JSON-RPC 2.0 over stdio, dual era: MCP 2026-07-28 (per-request _meta, server/discover, -32022) and legacy initialize (2025-11-25, 2025-06-18, 2025-03-26, 2024-11-05). Tools list_commands, execute_command (app.quit only with confirm: true), screenshot (image content), read_buffer, get_events (ring of 1000 events via events.subscribe); resources lutra://buffer/<path> for open buffers. Engine errors become tool errors with the engine code. Lines capped at 16 MiB, guard around every request, parser and tool args fuzzed. - lutra mcp [--port PORT] [--no-start] in apps/lutra-cli: connects through the rendezvous file and token, or starts Lutra. - Engine: read-only buffer.list command. - Tests: protocol units + proptest fuzz, integration over the token listener, stdio end to end. Docs: architecture (MCP), G7, ROADMAP. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
13 KiB
Lutra Roadmap
Stage: pre-alpha (M2 done) · next: M3 language server, terminal, CLI and MCP (the MVP); alpha when the six core workflows work end to end (see docs/roadmap.md › Alpha gate)
Last reviewed: 2026-10-15 · Last updated: 2026-10-19 · Change: minor (M3.8: MCP server) · Target: JetBrains PhpStorm 2026.x, described from public documentation only
Lutra is a clean-room, open-source PHP IDE with a Rust core and a Tauri + Vue interface, targeting the PhpStorm workflow. Agents can drive all of it over MCP, the lutra CLI and a JSON control channel. This page is the summary; the evidence is in docs/target-app-parity.md and the work list in docs/gaps.md.
Why pre-alpha: M0 built the foundation (workspace, quality gate, command registry with panic guard, control channel, a themed window). M1 lets you open a folder, browse it in the project tree, edit and save files in tabs, get unsaved text back after a crash, run every command from a palette and change settings and the theme in a settings dialog. M2 added PHP highlighting, find in files, a file structure view and go to file, class or symbol; PHP language intelligence (completion, diagnostics, go to definition) comes with M3. M1–M3 deliver the MVP.
Headline numbers
| Value | Kind | |
|---|---|---|
| Feature breadth (PhpStorm's actions, tool windows, languages, integrations exist) | ~15% (23 of 155 feature units of the public PhpStorm help have at least partial support; ~21% after the running cycle C4) | estimated |
| Feature depth (weighted by use, scored by behaviour) | ~3.2% (Editor core 10% × weight 12, Navigation 10% × weight 10, Project view 8% × weight 6, Search and replace 8% × weight 6, Settings 7% × weight 1, of 101) | estimated |
| To MVP (M3) | ~12 h left (M3 = 80 h raw × 0.15; done so far: M1 ~5.6 h, M2 ~6.1 h, measured); was ~190 h before calibration | estimated (factor per cycle: C1 0.07, C2 0.10, C3 0.14) |
| To full parity | ~450 h at the current factor of ~0.15 (range 300–3,300 h; the factor has risen each cycle, and deep areas such as type inference and inspections have no data yet) | estimated |
Hours are agent wall-clock hours for one agent. M0 took about 2 h of active agent time (measured from the session, excluding toolchain installation and waiting for owner decisions). Later milestones are estimated far more conservatively per line of code, because editor, language and UI work need integration, debugging and visual checks that the foundation did not. Estimates are recalibrated after each cycle from task start times and commit timestamps. Cycle C1 (M1.1–M1.6) took 2.5 h of agent time for 37 h estimated (factor 0.07); M1–M3 hours are scaled by ×0.1 since then, and the per-area hours below are the raw first estimates (multiply by ~0.1 for the calibrated value). PhpStorm is not inspected: the clean-room rule in AGENTS.md forbids it, so everything is measured from our code against public documentation.
By dimension
Dimension hours overlap: language, framework and tooling hours are part of the feature total.
| Dimension | % | One agent (h) | Doc |
|---|---|---|---|
| Features (depth) | ~3.2% | ~2,896 | target-app-parity.md |
| UI/UX fidelity | 1% (est.) | ~150 on top of features | ui-parity.md |
| PHP language support | 0% | ~920 (in features) | language-parity.md |
| Frameworks (Symfony, Laravel, Shopware, WordPress) | 0% | ~300 (in features) | framework-parity.md |
| Tooling (Git, terminal, debugger, tests, Composer, DB) | 0% | ~730 (in features) | tooling-parity.md |
| Localization (en, de) | 5% | ~38 | localization-parity.md |
| Performance | 0% | ~100 | gaps.md › G12 |
| Stability | 8% | ~95 | gaps.md › G11 |
| Agent automation | 26% (est.) | ~44 | AGENTS.md |
Stability 8%: no-panic lints, the engine guard, a fuzzed control protocol and a crash-recovery journal for unsaved buffers (property-tested: random edits, crash, restore) exist; the journal is not yet checked in the running app. Agent automation 26%: the control channel (ping, execute, listCommands, screenshot) works, and since M1.8 it also reaches the frontend's UI commands (opening the palette, toggling the project tree) through the uiCommand event; since M3.7 the lutra CLI opens folders and files at a line in the running instance (or starts it), --wait blocks until the tabs close (usable as core.editor), and a token-protected rendezvous file lets local tools find the instance; since M3.8 lutra mcp gives MCP clients such as Claude Code every command (with params docs and the enabled state), screenshots, the text of open buffers (also as resources) and the recent events, with app.quit only on explicit confirmation. Missing: change notifications, progress for long commands, the perf and diff subcommands. Both are estimated. Localization 5% (estimated): every UI string comes from an English catalog enforced by lint; engine messages, the language switch and German are missing. UI/UX fidelity 1% (estimated): light and dark themes switch live and pass a WCAG AA contrast test; the rest of the ui-parity checklist is open.
Features
| Area | % | One agent (h) |
|---|---|---|
| Editor core (buffers, undo, multi-caret, folding, soft wrap) | 10% (est.) | ~104 |
| Project view, VFS, file watching | 8% (est.) | ~40 |
| Navigation (go to class/file/symbol, recent files, structure) | 10% (est.) | ~102 |
| Search and replace (in file, in project, structural) | 8% (est.) | ~70 |
| Code completion | 0% | ~250 |
| Inspections and quick-fixes | 0% | ~300 |
| Refactoring | 0% | ~250 |
| Code style and formatting | 0% | ~120 |
| Version control (Git) | 0% | ~150 |
| Terminal | 0% | ~30 |
| Run configurations | 0% | ~50 |
| Debugger (Xdebug) | 0% | ~120 |
| Testing (PHPUnit, Pest) | 0% | ~80 |
| Composer and PHP interpreters | 0% | ~50 |
| Frameworks | 0% | ~300 |
| Database tools | 0% | ~250 |
| HTTP client | 0% | ~60 |
| Web languages (JS/TS, Vue, CSS, Twig, Blade) | 0% | ~200 |
| Docker and remote | 0% | ~120 |
| Settings, keymaps, themes | 7% (est.) | ~60 |
| Plug-ins and ecosystem | 0% | ~120 |
| AI features | 0% | ~60 |
All hours estimated. Weights, evidence and the missing items per area: target-app-parity.md.
Upcoming
- M3 language server, terminal and CLI (~12 h): completion, diagnostics, go to definition via Phpactor; terminal;
lutraCLI and MCP. MVP. - Then: Git (M4), debugger, tests and Composer (M5), own indexer and frameworks (M6).
Progress log
| Date | What landed |
|---|---|
| 2026-10-19 | M3.8: agents can drive Lutra over MCP: claude mcp add lutra -- lutra mcp, then Claude can list and run every command, look at the window, read open files including unsaved changes and follow what happens. Closing Lutra needs an explicit confirmation. |
| 2026-10-17 | M3.7: the lutra command: lutra src/Kernel.php:42 opens the file at line 42 in the running Lutra or starts it, lutra . opens a folder, lutra --wait FILE returns when the tab is closed (for git commit). Starting Lutra a second time hands the path to the first window. Only your user account can talk to the running instance. |
| 2026-10-17 | M2.6: go to any file, class or symbol by name: Mod+P, Mod+Alt+C and Mod+T open one popup with Files, Classes and Symbols tabs (Tab switches); results update as you type with the matched letters marked, library code is labelled, a hint shows while the index is still building, and Enter jumps there. Agents can open it with a query. |
| 2026-10-17 | M2.7: a Structure tool window next to Project (new left stripe, Mod+2) shows the classes, functions, methods, properties and constants of the active PHP file as a tree that follows tabs and edits; click or Enter jumps to the declaration. The file is reparsed in the background, so big files never stall typing. |
| 2026-10-17 | M2.4: Find in files has a window: Mod+Alt+F opens a bottom panel with the selection or the word at the cursor as the query; results stream in grouped by file with the hit highlighted, with match case, whole words, regex and file globs; Enter or a click jumps to the match; a new search cancels the old one. |
| 2026-10-16 | M2.2: PHP files are highlighted (with the HTML around <?php blocks), and so are HTML, CSS, JS/TS, JSON, YAML and Markdown, in Lutra's own colours for both themes, each readable at 4.5:1 or better; editor.reveal jumps to a file position for the coming Find, Go-to and Structure windows and for agents. |
| 2026-10-16 | M1.9b, M1 done: a settings dialog (Mod+,) built from the engine's settings, with search, a control per type, reset to default and inline errors; changes apply at once and follow edits from agents live; the theme picker switches at once and the editor font size applies to the editor. |
| 2026-10-16 | M1.8: a command palette (Mod+Shift+P) finds every command by fuzzy match, recent ones first, with shortcuts shown and unavailable ones dimmed; one global keymap serves engine and UI commands (Mod+1 toggles the project tree); agents can trigger UI commands over the control channel. |
| 2026-10-16 | M1.12: errors and notices reach the user as toasts instead of stderr: a corrupt settings file or recovery entry, a folder that cannot be opened (also with unsaved changes when switching projects), stopped file watching and failed commands without their own UI. Info toasts go away after 5 s, warnings and errors stay. |
| 2026-10-16 | M1.7: unsaved text survives a crash. Dirty buffers are journaled about a second after the last edit; the next open offers them in a restore dialog (restore all, selected, or discard), with line endings and BOM kept. |
| 2026-10-16 | M1.10: Lutra Light next to Lutra Dark, switched live through the appearance.theme setting or following the OS (system); the window starts in the stored theme; the editor, tabs and notices get their own tokens; both themes pass a WCAG AA contrast test. |
| 2026-10-15 | M1.11: every interface string comes from an English message catalog (ui/src/i18n/en.json), enforced by a lint rule and a catalog test; the groundwork for German. |
| 2026-10-15 | M1.6: files open in tabs with a CodeMirror editor synced to the engine; save, save all, close with a Save / Don't Save / Cancel question, a notice when a file with unsaved changes changes on disk; a 1 MB file opens in about 90 ms. |
| 2026-10-15 | M1 so far: text buffer, VFS with native watching, engine workspace (files, buffers, edit protocol, events), desktop bridge: open a folder by picker or command line, browse it in the project tree with live updates from disk. |
| 2026-10-14 | M0 foundation: Cargo workspace with layered crates, no-panic lints with a self-test, cargo xtask ci (fmt, clippy, tests, layers, assets, clean-room, licences, frontend lint/typecheck/tests), Vue frontend, command registry with panic guard (app.about, app.quit), JSON control channel, Tauri window themed from Rust tokens, CI workflow. First estimates. |
| 2026-10-13 | Project started: name, conventions, agent rules and progress docs. |
Revision history
| Date | Change | Summary |
|---|---|---|
| 2026-10-18 | minor | Feature breadth measured against the full PhpStorm help inventory (155 units); M7–M9 added |
| 2026-10-19 | minor | M3.8: MCP server; agent automation 18% → 26% (est.) |
| 2026-10-17 | minor | M3.7: lutra CLI and single instance; agent automation 10% → 18% (est.) |
| 2026-10-17 | minor | Cycle C3 review: M2 done (~6.1 h measured), factor 0.14, MVP ~12 h left |
| 2026-10-17 | minor | M2.6: Navigation 4% → 10%, feature depth ~3.2% |
| 2026-10-17 | minor | M2.7: Navigation 4%, feature depth ~2.6% |
| 2026-10-17 | minor | M2.4: Search and replace 8%, feature depth ~2.2% |
| 2026-10-16 | minor | M2.2: Editor core 10%, feature depth ~1.7% |
| 2026-10-16 | minor | Cycle C2 review: M1 measured at ~5.6 h, C2 factor 0.10 confirms ×0.1 |
| 2026-10-16 | minor | M1.9b: M1 done, stage line, Settings row 7% (est.), progress log |
| 2026-10-16 | minor | M1.8: progress log, agent automation note (scores unchanged) |
| 2026-10-16 | minor | M1.12: progress log (headline unchanged) |
| 2026-10-16 | minor | M1.7: Stability 8%, Editor core 9%, feature depth ~1.6% (with Settings 4% from M1.10) |
| 2026-10-16 | minor | M1.10: UI/UX fidelity 1%, Settings row 4% (estimates), progress log |
| 2026-10-15 | minor | M1.11: localization 5% (English UI catalog), progress log |
| 2026-10-15 | minor | Cycle C1 review: stage line, breadth estimate, hours recalibrated ×0.1 |
| 2026-10-15 | minor | Feature depth ~1.4%: tabs and editor (M1.6) |
| 2026-10-15 | minor | Feature depth above 0%: project tree runs against the engine (M1.4) |
| 2026-10-14 | minor | M0 done; MVP and parity estimates; stability and agent automation scored |
| 2026-10-13 | major | Initial roadmap |