mcp-firstlisted

@markdownai # MCP-first (fleet doctrine) > **Read when:** boot step 14, before inventing scripts, before scraping GitHub for "operator voice," before `Grep`/`Bash` on fleet stores. > > **Not in scope:** full tool schema — use `fleet_tool_guide` or `sap/mcp_registry.json` when you need depth. --- ## How you log in (orient) Run **in parallel** at session start (Phase 2 boot): | Call | What you get | |------|----------------| | `boot_digest(app_id=<agent>, workspace=<repo>)` | Verified handoff claims, next bite, **`tools:` line = wired MCP servers** | | `willow_status` or `fleet_status(app_id=<agent>)` | Postgres, SOIL, Kart queue, watchmen | | `handoff_latest(app_id=<agent>, workspace=<repo>)` | Session continuity (cold boot / `fast_path: no`) | **`app_id`** = fleet identity (`WILLOW_AGENT_NAME` / `.willow/active-agent`) — not persona, not dispatch target. Postgres down in **private-config** → hard stop. Grove down → degraded, continue. --- ## Which MCPs exist (manifest) **Do not guess.** Read what this workspace actually wired: 1. **`boot_digest`** → `digest.sections.mcp_inventory.mcp_servers` (e.g. `willow`, `codebase-memory-mcp`) 2. **IDE** → `.cursor/mcp.json` / `.mcp.json` in repo root 3. **`fleet_tool_guide(app_id=<agent>)`** → grouped verbs for active `WILLOW_MCP_PROFILE` Typical Willow 2.0 desk: | Server | Role | Reach via | |--------|------|-----------| | **willow** | Fleet memory, SOIL, KB, sessions, Kart, handoffs, facades | `willow_*`, `kb_*`, `soil_*`