MCP Discovery

ToolYour MCP uses progressive disclosure so agents never load hundreds of tool schemas at once.

Start with `plan_task` + `solve_task`

plan_task(goal) — free ranked plan + credit estimate
run_playbook or solve_task(goal) — jobReport + loop.remainingFixes + loop.gate
Host applies fixes in the repo, then verify_task until loop.gate is pass

invoke_tool is advanced (one-off operationId), not the default for ship/SEO/security jobs.

If the server cannot auto-route, the response includes taskSuggestions / toolSuggestions (empty when out of catalog).

Meta-tools

Tool	Free?	Purpose
`plan_task`	Yes	Plan without execute
`solve_task`	Suggest free; run bills	Primary entry
`run_playbook`	Bills	Skill → workflow
`verify_task`	Bills	Delta vs baseline (optional async)
`discover_tools`	Yes	Keyword search
`list_categories`	Yes	Tool families
`get_tool_schema`	Yes	One schema
`fetch_payload`	Yes	Full `dataRef` body
`get_run`	Yes	Poll async `runId` (read `resultStatus`)
`list_skills` / `load_skill`	Yes	Playbooks

Search tips

Be specific: docx pdf beats convert
Keep limit low (default 10)
After discovery: get_tool_schema → invoke_tool (advanced one-off only)

What is excluded

Only tools with hasApi: true appear in MCP.

Recommended patterns

plan_task(goal) → solve_task(goal, responseMode: "compact")
run_playbook("ship-gate", { url })
list_categories() → discover_tools("pdf", category: "Documents") → get_tool_schema → invoke_tool

See Playbook catalog and Calling tools.