Etchplan MCP server — let an agent invoke Etchplan (mine / compile / replay repeated workflows). Compile an AI agent's repeated workflows into deterministic, auditable routines that replay at $0, with a fallback to the agent.
Six tools with complete input schemas and descriptions, but quality is uneven. Tool names follow verb_noun convention (etchplan_status, etchplan_mine, etchplan_compile, etchplan_run_meta_tool, etchplan_list_meta_tools). Descriptions range 50 - 150 chars, meeting baseline minimums (p10=34, p90=392). All parameters have type and description. However, descriptions lack actionable context (WHEN to use, prerequisites, error recovery). Output schemas are not documented in source. No parameter constraints (enums, ranges, patterns) visible. Error handling guidance absent. etchplan_run_meta_tool accepts 'inputs' as bare object with no schema, forces LLM to guess structure. etchplan_mine has 'persist' default=true with no warning about side effects (writes to DB). No idempotency guarantees stated.
Compile a mined pattern into a guarded execution plan; returns a summary.
List the compiled meta-tools available as callable bundles (id, bundled steps, params, approval).
Mine repeated agent workflows from stored traces; returns a compact summary.
Per-plan ROI/lifecycle stats (invocations, fallback rate, drift status).
Run any compiled meta-tool plan by path with inputs (one deterministic call over a bundled sub-sequence).
Counts of stored traces/patterns/plans for an Etchplan database.
etchplan_run_meta_tool 'inputs' parameter is untyped object with no schema. LLM cannot infer required fields, types, or constraints. Forces guessing at structure.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields (e.g., pattern_id from etchplan_mine to feed etchplan_compile).
etchplan_mine 'persist' defaults to true (writes to DB) with no warning in description. Agents may not realize side effect; no dry-run option offered.
Descriptions lack actionable context: no WHEN to use, no prerequisites, no error recovery guidance. E.g., 'Mine repeated agent workflows' does not explain when mining is useful or what happens if DB is empty.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 63 | 2025-06-18+ | v2 |
No parameter constraints visible (enums, min/max, patterns). E.g., min_support has no range guidance; registry_dir has no validation. Invites invalid LLM input.