# Robisie Planer — MCP Gateway Skill **Version:** 1.0.0 **Gateway URL:** https://robisie.app/api/mcp **Auth:** Bearer token (generate at https://robisie.app/dashboard/connect) --- ## What is this Robisie Planer is an agent-native project management board. Your operator manages their projects through a kanban UI; you (the agent) connect via MCP and read + write tasks directly — no human hand-off required. **You are the primary user of this board.** The operator sets priorities, you execute. --- ## How to connect ### Claude Code (fastest) ``` claude mcp add --transport http robisie-planer https://robisie.app/api/mcp --header "Authorization: Bearer YOUR_KEY" ``` Add `-s user` to make it available across all projects. ### Cursor / Windsurf — add to mcp.json / mcp_config.json: ```json { "mcpServers": { "robisie-planer": { "url": "https://robisie.app/api/mcp", "headers": { "Authorization": "Bearer YOUR_KEY" } } } } ``` ### Claude Desktop — add to claude_desktop_config.json: ```json { "mcpServers": { "robisie-planer": { "command": "npx", "args": ["-y", "mcp-remote", "https://robisie.app/api/mcp", "--header", "Authorization:${AUTH_TOKEN}"], "env": { "AUTH_TOKEN": "Bearer YOUR_KEY" } } } } ``` The operator gets their key at https://robisie.app/dashboard/connect. Keys are one-time-display and hashed at rest. --- ## Data model ``` Project └── Track (column / swim lane) └── Task (card) ├── status: spec | plan | w_pracy | do_akceptacji | zrobione ├── dependsOn: [taskId, ...] — blocks this task from get_next_task ├── specRef: string — pointer to spec doc / ticket ├── assignee: string — who/what is doing this ├── rob: string — freetext notes └── details: Record — structured metadata ``` **Status flow:** `spec` → `plan` → `w_pracy` → `do_akceptacji` → `zrobione` A task is "unblocked" when all its `dependsOn` tasks are `zrobione`. --- ## Available tools (19) ### Board overview - **get_board(projectId)** — full board: all tracks + task digest (id, title, status, position, rob). Excludes `zrobione` by default; pass `statuses: ["zrobione"]` to include. Also returns `robProgress` per-ROB rollup. Start here. - **get_next_task(projectId)** — returns the highest-priority unblocked task in `spec` or `plan` status (priority order: goal → track order → position). Returns `null` when nothing is unblocked. ### Projects - **list_projects()** — list all your projects (zero args) - **create_project(name)** — create a project; returns new project with id - **update_project(id, { name? })** — rename the project (slug updates automatically) - **delete_project(id)** — delete project and all its content (cascades) ### Tracks (columns) - **list_tracks(projectId)** — list tracks in position order - **create_track(projectId, name, color?)** — add a column (auto-appended to end; `color` is optional) - **update_track(id, { name?, position?, color? })** — rename, reorder, or recolor - **delete_track(id)** — fails with isError if track still has tasks (move them first) ### Tasks (cards) - **list_tasks(projectId, { trackId?, statuses?, full? })** — list tasks sorted by position. Default omits `zrobione`; `statuses` = explicit status array override (e.g. `["zrobione"]`); `full: true` returns all fields plus `depsReady` boolean (whether all dependsOn are `zrobione`) - **get_task(id)** — full task detail including `details`, `assignee`, `dependsOn` - **create_task(projectId, trackId, title, { status?, details?, specRef?, assignee?, rob?, dependsOn? })** — create a card; returns slim echo + `similarOpenCards` anti-duplicate warning - **create_tasks(projectId, tasks[])** — batch create up to 200 cards (sequential; per-card errors don't stop batch); `dependsOn` must reference pre-existing cards (no intra-batch forward refs) - **update_task(id, { title?, status?, details?, specRef?, assignee?, rob?, dependsOn? })** — update fields. ⚠️ `details` as object REPLACES the entire field — use `append_detail` to add fields without overwriting - **move_task(id, { trackId?, afterId?, beforeId? })** — move card to a different track and/or reorder. `afterId` = card to sit after; `beforeId` = card to sit before (both must be in the target track). Omit both → append to end of track - **delete_task(id)** — delete a card - **reserve_task(id)** — atomically claim a task: if `assignee` is empty, sets `assignee='pending'` AND `status='w_pracy'` in one step. Returns the card on success; returns `null` if already claimed. No separate status update needed after reserve - **append_detail(id, patch)** — shallow-merge a flat JSON object into `details` without overwriting other keys (`patch` = flat object, top-level keys only). Use instead of `update_task` when adding a single field (e.g. evidence, PR link, timestamp) --- ## Typical agent workflow ``` 1. list_projects() → find your project id 2. get_next_task(projectId) → find the next unblocked task 3. reserve_task(taskId) → claim it atomically (sets assignee='pending' + status='w_pracy') 4. [do the work] 5. update_task(taskId, { status: "do_akceptacji", rob: "completed:

" }) OR update_task(taskId, { status: "zrobione" }) → if no operator review needed 6. append_detail(taskId, { completed_at: new Date().toISOString() }) 7. goto 2 ``` --- ## Important behaviors - **RLS isolation:** You only see and modify the data of the authenticated user (Bearer token owner). There is no cross-tenant access. - **get_next_task priority:** Returns the task with lowest position among unblocked tasks in `spec` or `plan` status. Tasks whose `dependsOn` are not all `zrobione` are excluded. - **reserve_task:** Atomically sets `assignee='pending'` and `status='w_pracy'` in one step — no separate status update needed. If two agents call simultaneously, one gets the card, the other gets `null`. - **append_detail shallow merge:** Only top-level keys are merged. Nested objects under existing keys are replaced, not deep-merged. Use flat objects. - **create_tasks batch:** Prefer over multiple `create_task` calls when seeding a project. Up to 200 tasks per call; `dependsOn` must reference cards already in the DB. --- ## Getting started (new project) ``` # 1. Create project create_project("My AI Project") # 2. Set up tracks create_track(projectId, "Backlog") create_track(projectId, "W toku") create_track(projectId, "Gotowe") # 3. Seed tasks (batch) create_tasks(projectId, [ { trackId: backlogId, title: "Research approach", status: "plan" }, { trackId: backlogId, title: "Implement feature A", status: "spec" }, ]) # 4. Start the loop get_next_task(projectId) → reserve_task → work → update_task → repeat ``` --- ## Troubleshooting - **401 Unauthorized** → Key is invalid or revoked. Operator should generate a new key at https://robisie.app/dashboard/connect. - **get_next_task returns null** → All tasks are either `w_pracy`, `do_akceptacji`, `zrobione`, or blocked by unresolved `dependsOn`. Ask operator to review. - **Task not visible** → Check that you are using the correct `projectId`. Call `list_projects()` to verify. - **reserve_task returns null** → Task already claimed by another agent. Call `get_next_task` again for the next available card. --- *This skill file is served at https://robisie.app/SKILL.md for agent self-onboarding.*