Memory and workflow automation for MCP, Codex, and Claude Code. Reuse repeatable developer operations through skills and explicit feedback.
FlowMind MCP server has 21 tools with visible schemas and descriptions, but quality is inconsistent. Core tools (flowmind_process, flowmind_start_task, flowmind_task_status) have adequate descriptions (50-100 chars) and proper input schemas with typed parameters. However, 6 skill-execution tools (flowmind_skill_*) have generic descriptions ('Execute [skill-name] skill | Capabilities: [category]') that lack actionable context for LLM selection, they do not explain WHEN to use each skill, what distinguishes them, or what they return. Parameter descriptions are minimal ('Input for the skill', 'Optional context'). Output schemas are entirely undocumented across all tools, LLMs cannot plan downstream calls or extract structured data. Error handling is absent: no recovery guidance, no categorization of retryable vs fatal errors, no invalid-value feedback. Naming is verb-prefixed (good), but skill tools use a repetitive 'flowmind_skill_*' pattern that obscures semantic differences. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk classifications in the spec.
Get AI model status and configuration
Cancel a queued asynchronous FlowMind task
List FlowMind capability groups, mapped skills, and discovery metadata
Execute an allowlisted workflow after approval-token confirmation
Get detailed information about a specific skill
View sanitized learned skill feedback, scenes, bindings, and local material package metadata
Get learning statistics
Skill-execution tools (flowmind_skill_*) have generic, non-actionable descriptions. 'Execute [skill-name] skill | Capabilities: [category]' does not explain WHEN to use each skill, what distinguishes it from similar tools, or what it returns. LLMs cannot reliably select the correct skill.
No output schemas documented for any tool. LLMs cannot plan downstream calls, extract structured data, or validate responses. This violates the pattern:tool requirement and forces agents to guess at response structure.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 47 | 2025-06-18+ | v2 |
List all available FlowMind skills
List recent asynchronous FlowMind tasks
Dry-run a guarded follow-up workflow and return its approval token
Process a request using FlowMind AI agent. This is the main entry point for using FlowMind.
Resume a previously recorded workflow execution from the first unfinished step
Execute api-sync skill | Capabilities: docs
Execute archive-change skill | Capabilities: reporting
Execute auto-flow skill | Capabilities: workflow
Execute build-report skill | Capabilities: reporting
Execute change-impact-review skill | Capabilities: review
Execute code-review-audit skill | Capabilities: review
Start an asynchronous FlowMind tool execution and return a pollable task id
Get task status, progress, and result for an asynchronous FlowMind execution
List recent recorded workflow executions and their resume state
No error handling guidance. Tools do not indicate which errors are retryable, what the LLM should do on failure, or how to self-correct invalid input. E.g., flowmind_start_task does not explain what happens if 'tool' parameter is invalid.
Skill-execution tool parameters lack meaningful descriptions. 'Input for the skill' and 'Optional context' are too generic, LLMs cannot infer what data structure or content is expected. No examples, constraints, or format guidance.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk classifications in the spec. Tools marked WRITE (flowmind_start_task, flowmind_execute_workflow, flowmind_resume_workflow, flowmind_skill_archive-change, flowmind_skill_auto-flow) should declare destructiveHint=true to guide agent planning.