📚 Model Context Protocol server for AI-powered storytelling — Mage image generation and content tools.
The Echoes MCP server demonstrates moderate definition quality with clear naming conventions and reasonable descriptions, but has systematic gaps in parameter documentation and output schema clarity. All 11 tools follow verb_noun naming patterns (words-count, stats, index, search, list, graph-export, history, review-generate, review-status, review-apply, timeline-overview), which meets baseline expectations. Tool descriptions average 90-120 characters and are reasonably specific about function. However, parameter descriptions are inconsistent, some params like 'arc' and 'dbPath' repeat across tools without variation, suggesting copy-paste rather than careful crafting. No output schemas are documented in the provided source, which prevents LLMs from understanding what fields to expect and planning downstream calls. Error handling is basic (try/catch with formatError), offering no recovery guidance. The server does not validate inputs (e.g., checking enum values for graph-export format) inside tool implementations.
List configured Mage characters with their placeholders.
Queue a single Mage image/video generation.
Queue multiple Mage image generations (one prompt per line).
Cancel a job from the Mage queue.
List items currently in the Mage generation queue.
Pause the Mage queue (current job continues to completion).
Resume the Mage queue processing.
No output schemas documented. LLMs cannot determine what fields each tool returns, forcing them to guess about downstream usage and preventing automatic response validation.
Parameter descriptions are generic and repeated across tools without clarification of context-specific behavior. E.g., 'arc' description is identical on all tools but behavior differs (some filter, some require arc). LLMs cannot distinguish when arc is optional vs. required.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 14 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 35 | - | v1 |
List generated Mage results. Returns up to 1000 results (no pagination). If truncated, save some results to free up space.
Save a generated image: call saveMageResult mutation, then download from Mage CDN to local filesystem.
Save all unsaved generated results for the specified arcs to local filesystem.
Get overall Mage system status: queue, results, deployment, auth.
Get the MCP server version.
Counts the number of words in a given text file.
No input validation visible in tool implementations. graph-export accepts format enum (json, cypher, csv) but no enforcement shown. search accepts limit but no bounds validation. LLMs can pass invalid values without actionable error feedback.
Error handling is minimal. formatError() returns raw Error.message without categorizing as retryable/user-fixable/fatal or providing recovery guidance. E.g., 'Index creation failed' gives LLM no direction on next steps.
No confirmation or dry-run support for destructive tools (index with force=true, review-apply). Agents can irreversibly re-index or apply corrections without safeguards.
Several tools accept optional numeric parameters (limit in search/history) with no documented bounds. LLMs could pass extreme values (limit: 999999) causing performance issues.
review-status accepts optional 'file' parameter but description does not clarify format (YAML path?) or what happens if missing. Ambiguous parameter design.