Stdio MCP server exposing the Consensus Validation Protocol as a single `consensus` tool. Thin wrapper over ai-consensus-core.
The server defines 17 tools with mixed quality. The primary `consensus` tool has a well-structured JSON schema with 11 optional parameters, clear descriptions, and proper type constraints (e.g., maxRounds 1-10, participantTemperature 0-2). However, 16 preset-specific tools (consensus_code_review, consensus_architecture_debate, etc.) lack visible input schemas in the source code, only tool names and brief descriptions are shown. Descriptions are present but generic (e.g., 'Expert-panel preset tool for code review consensus'). The consensus tool description is strong (~180 chars), but preset tools average ~80 chars and lack actionable guidance on when to use each. No parameter descriptions visible for preset tools. Error handling and output schemas are not documented in the provided code.
The generic consensus tool that takes a fully free-form prompt with every engine knob exposed. Runs the Consensus Validation Protocol with configurable participants, rounds, and convergence parameters.
Expert-panel preset tool for architecture debate consensus. Applies the architecture_debate preset's persona panel and tuned defaults.
Expert-panel preset tool for architecture_v2 consensus. Applies the architecture_v2 preset's persona panel and tuned defaults.
Expert-panel preset tool for code review consensus. Applies the code_review preset's persona panel and tuned defaults.
Expert-panel preset tool for code_review_v2 consensus. Applies the code_review_v2 preset's persona panel and tuned defaults.
Expert-panel preset tool for debug postmortem consensus. Applies the debug_postmortem preset's persona panel and tuned defaults.
16 of 17 preset tools lack visible input schemas in source code. These tools are inferred to exist but not explicitly defined in the provided code.
Preset tool descriptions are generic (~80 chars) and lack actionable guidance. E.g., 'Expert-panel preset tool for code review consensus' does not explain when to use this vs consensus_code_review_v2, what panel it applies, or what output to expect.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 41 | 2026-07-28+ | v2 |
Expert-panel preset tool for decision making consensus. Applies the decision_making preset's persona panel and tuned defaults.
Expert-panel preset tool for decision_making_v2 consensus. Applies the decision_making_v2 preset's persona panel and tuned defaults.
Expert-panel preset tool for incident_postmortem_v2 consensus. Applies the incident_postmortem_v2 preset's persona panel and tuned defaults.
Expert-panel preset tool for ml_research_2026 consensus. Applies the ml_research_2026 preset's persona panel and tuned defaults.
Expert-panel preset tool for product_strategy consensus. Applies the product_strategy preset's persona panel and tuned defaults.
Access project-scoped memory context. Only available when memory is enabled in configuration.
Recall previous consensus decisions from the memory layer. Only available when memory is enabled in configuration.
Expert-panel preset tool for research synthesis consensus. Applies the research_synthesis preset's persona panel and tuned defaults.
Expert-panel preset tool for research_synthesis_v2 consensus. Applies the research_synthesis_v2 preset's persona panel and tuned defaults.
Expert-panel preset tool for security_redteam consensus. Applies the security_redteam preset's persona panel and tuned defaults.
Query what decisions have been made in the current project. Only available when memory is enabled in configuration.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields without knowing the response structure. Baseline: 100% of A+ tools document return types.
Memory-dependent tools (consensus_recall, consensus_project_memory, consensus_what_we_decided) lack descriptions explaining prerequisites ('Only available when memory is enabled') and expected input/output. LLMs cannot determine if memory is enabled or how to invoke these tools.
No error handling guidance documented. If consensus fails to converge, a participant is unavailable, or memory is not enabled, the LLM has no recovery path. Baseline: error responses must guide the LLM on retry, user-fixable, or fatal classification.