A workflow composition and execution planning service that builds execution plans from workflow step definitions with cost estimation, retry logic, and timeout configuration
This MCP server has critical definition quality gaps. The single tool 'compose' lacks proper MCP tool registration semantics, has a vague description that doesn't explain WHEN to use it or its operational implications, and the input schema is visible but severely underspecified. No parameter descriptions are provided in the schema, violating pattern:tool-description baseline of 100% of A+ tools having parameter annotations. The tool description (78 chars) falls short of the 194-char baseline and fails to state whether this operation modifies state or is read-only (critical for idempotence and retry semantics). No output schema is documented despite the tool returning a complex nested structure. The tool also lacks error handling guidance, recovery instructions, and actionable feedback for LLM use.
Composes a workflow execution plan from a list of workflow steps and a trigger event. Generates a workflow ID, calculates step costs and timeouts, and returns an execution plan with estimated duration.
Parameter descriptions missing. Input schema defines 'workflow_steps' and 'trigger' but provides no descriptions of what each parameter controls, valid formats, or constraints. Rubric baseline: 100% of A+ tools have param descriptions.
No output schema documented. Tool returns complex nested object with execution_plan (list of dicts), workflow_id, and estimated_duration_s, but the response structure is not declared. LLMs cannot plan downstream calls or extract chaining IDs without knowing what fields are returned.
Tool description lacks critical context. At 78 chars, it is below the 194-char baseline and does not explain: (1) When to call this tool vs alternatives, (2) whether this operation is idempotent/retryable, (3) what the workflow_id can be used for downstream, (4) error conditions or failure modes.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 31 | 2026-07-28+ | v2 |
No error handling or recovery guidance. If workflow_steps is empty, malformed, or contains an unknown action, the tool returns silently or produces unexpected output. No error schema, no guidance for LLM recovery (e.g., 'unknown action, valid actions are: scrape_web, call_api, ...').
Parameter type constraints not documented in descriptions. 'workflow_steps' is typed as list but the schema does not constrain item structure or validate 'action' enum values. LLMs may pass unknown actions (e.g., 'unknown_action') and expect silent fallback to 'run_code', leading to bugs.
Tool naming ('compose') is generic and does not follow verb_noun convention. 'compose' alone is vague, does it validate? Generate? Schedule? Rename to 'compose_workflow' or 'create_workflow_plan' for clarity.
No idempotence guarantee stated. The tool generates a workflow_id using time.time() and hashlib, making each call produce a unique ID even with identical inputs. This is non-idempotent and will cause agents retrying on transient failures to create duplicate workflows.