Deterministic governance lint for MCP server configurations, plus reference tools for the 7 rules. Audit MCP configs, enforce behavioral guardrails, track audit trails, and provide governance framework instructions.
The Nervous System is a governance/audit-focused MCP server with 12 tools. All tools have names and descriptions, but quality is uneven. 11 tools are READ_ONLY metadata/instruction getters (get_*, check_preflight) with minimal practical value to agents, they return static text or configuration guidance rather than actionable data. The one genuinely functional tool (audit_mcp_config) has a proper schema. Descriptions range from adequate (audit_mcp_config: 250+ chars) to vague (get_framework, get_nervous_system_info: 50-80 chars). Parameters that do exist (file_path in check_preflight, config/config_json in audit_mcp_config) are described, but the majority of tools have no parameters at all, they are simple information endpoints. Error handling is absent, tools return static strings with no guidance for LLM recovery. The server reads well as governance documentation but fails the agent tooling bar: agents cannot *do* anything with these tools except read instructions. Per-tool analysis shows 10 tools at ~30-40 (naming is clear, descriptions present but thin, schemas empty), and 2 tools (check_preflight, audit_mcp_config) at 50-60 (slightly better schemas and descriptions but still gaps).
Governance lint for MCP server configurations (free, local). Checks: plaintext secrets in env, unpinned packages, auto-install flags, broad filesystem scopes, shell wrappers, non-TLS remote transports, unpinned docker tags. Returns findings with severities and a 0-100 score.
Check if a file is protected (UNTOUCHABLE)
Returns how to verify the hash-chained audit trail. This tool does NOT verify anything; verification needs the chain file, which is server-side.
Returns a ready-to-run background dispatch command. This tool does NOT dispatch anything; you run the command.
Get a task file template for dispatching agents
Get the complete Nervous System governance framework with all 7 rules
10 of 12 tools have empty input schemas (type: object with no properties defined). This cascades to low overall scores.
11 of 12 tools are purely informational (get_* static instruction/metadata getters). They do not enable agents to perform actions, only to read governance text. This violates the composition pattern: tools should perform work or retrieve actionable data, not just echo documentation.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 46 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 33 | - | v1 |
Get a session handoff template
Returns the command to stop the fleet. This tool does NOT stop anything; it prints instructions. Enforcement is server-side.
Get system info, version, and production stats
Get the origin story of The Nervous System
Get a step-back reflection prompt
Get the worklog entry format
Tool descriptions are brief (30 - 75 chars for most get_* tools), below the 50-char minimum for clarity and below the 194-char baseline. Descriptions lack context on WHEN to call the tool, what it returns, or why an agent would need it. E.g., 'Get the origin story of The Nervous System' tells an LLM nothing about when to invoke it.
No error handling or recovery guidance. Tools return static strings with no indication of what to do if a call fails or if data is unavailable. Error responses must guide the LLM on retry or next steps (pattern: recovery-guide).
check_preflight and audit_mcp_config expect file paths or config objects as input, but error messages for invalid paths (e.g., nonexistent files, malformed JSON) are not documented. LLMs cannot self-correct without clear error guidance.