Multi-tool MCP server for specialized AI agents - each agent as its own tool. Exposes each specialized AI agent as an individual MCP tool, providing direct access to Claude Code CLI with custom configurations.
This server exposes 8 specialized agent tools via FastMCP. While the tool names are clear and descriptions are present, there are critical gaps in schema definition, parameter documentation, output structure, and error handling that prevent higher scoring. All 8 tools share identical input schemas (prompt + session_reset), but lack comprehensive parameter descriptions, output schemas, and error guidance. The descriptions are domain-specific and reasonable (100-150 chars each), but do not guide LLM behavior on when to use one agent vs. another or what the expected output structure is. No tool annotations (readOnlyHint/destructiveHint) are declared despite clear risk profiles (7 READ_ONLY, 1 WRITE). Error handling is absent, there is no documentation of failure modes, recovery paths, or what the LLM should do if a tool call fails.
Market research and competitive analysis, requirements gathering and documentation, brainstorming and ideation facilitation, project discovery for greenfield and brownfield projects, creating comprehensive project briefs, strategic research planning
System architecture design, technology stack selection, API and database schema design, infrastructure planning, integration architecture, performance and security considerations
Full-stack code implementation, feature development from stories, API and backend development, frontend implementation, database operations, code debugging and testing
Creating Product Requirements Documents (PRDs), feature definition and prioritization, success metrics and KPI definition, product vision and roadmap planning, stakeholder alignment documentation, epic and feature breakdown
PRD and architecture validation, document sharding into manageable epics, backlog prioritization, requirement quality assurance, sprint readiness assessment, stakeholder requirement validation
Code review and quality assurance, refactoring for better maintainability, test coverage improvement, security vulnerability identification, performance optimization, best practices enforcement
Output schemas are not documented. No visible schema definition for what each tool returns, LLMs cannot plan downstream calls or extract structured data.
Parameter descriptions are minimal. 'prompt' and 'session_reset' are named but lack guidance on format, length, or constraints. No hints on what kind of prompts each agent expects (e.g., analyst expects market research briefs; pm expects feature requests).
No tool annotations for risk profile. All 8 tools are registered without readOnlyHint/destructiveHint. The 'dev' tool performs WRITE operations and should be marked as such so agents can reason about rollback and safety.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 48 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 30 | - | v1 |
Converting epics to detailed user stories, task breakdown and estimation, acceptance criteria definition, sprint planning preparation, story implementation readiness, developer-ready documentation
UI/UX design specifications, wireframe and mockup creation, design system development, user flow optimization, accessibility planning, generating prompts for AI UI tools (v0, Lovable)
No error handling guidance. Tool descriptions do not explain what happens on failure, how to retry, or what the LLM should do next. A tool that fails with 'Agent returned error' leaves the LLM with no recovery path.
Tool descriptions do not disambiguate when to use one agent vs. another. All are described in terms of their domain, but lack guidance like 'Use analyst first to gather requirements; then pass results to pm to create a PRD' or 'Call dev only after architect defines the schema.'
Tool naming lacks clear verb-noun pattern for some tools. While 'analyst', 'pm', 'ux_expert' are clear role names, they do not follow the verb_noun pattern (e.g., 'analyze_market', 'define_product', 'design_ux') that would immediately convey action to an LLM.
qa tool definition is in a different file (server.py) from the other 7 (resource_manager.py). This inconsistency, combined with visible source code truncation in the qa section, suggests the tool may be incompletely defined. Its schema and full definition cannot be verified from the provided source.
Composition issue: All 8 tools accept identical input ('prompt' + 'session_reset'). There is no indication of how tools chain together or what data flows between them. If 'analyst' output should feed into 'pm', the schema must document this.