Crewship exposes 2 tools with partial schema coverage. integration_tools_refresh has a well-structured input schema with nested object definitions for tool array items, but lacks output schema documentation. inspect_run has minimal schema (only run_id string param) and no output schema. Both tools have descriptions (60-80 chars), meeting the minimum but falling short of LLM-optimized length (50-200 chars baseline). Parameter descriptions are present but generic. No error handling guidance, no tool annotations, no pagination support. The server is HTTP-based (good for transport) but tool definitions lack the depth needed for confident agent reasoning.

• Add output schema to both tools. For inspect_run, document fields like run_id, status, timeline (array of events), tool_calls (array with name, input, output), errors (array), execution_time_ms. For integration_tools_refresh, document success response with integration_id, updated_tool_count, validation_errors (if any).
• Expand tool descriptions to 150-250 characters. Example for inspect_run: 'Retrieve detailed execution trace for a pipeline run, including timeline of steps, tool invocations with inputs/outputs, errors encountered, and total execution time. Use this to debug failed runs or audit what a pipeline did.'
• Add format constraint to inspect_run's run_id parameter. Update schema to include pattern: '^(r_|prn_)[a-zA-Z0-9]+$' and description: 'Run ID in r_ or prn_ namespace (e.g. r_abc123def456).'
• Document error responses with recovery guidance. E.g. 'If tool schema validation fails, return {error: "Invalid tool schema", invalid_tools: [{name, reason}], suggestion: "Fix schema and retry with corrected definitions"}.'
• Add limit and validation to integration_tools_refresh. Constrain tools array to max 100 items, add description: 'Array of tool definitions (max 100). Each must have unique name, non-empty description, and valid JSON schema if provided.'