Loading ToolBench

Reading the current corpus…

mcp-recon MCP server · F · ToolBench

mcp-recon

github.com/euanmcrosson-dotcom/mcp-recon

Reverse-engineer MCP server tool surfaces. Enumerate, fuzz, classify, report. CLI + library that discovers MCP tool surface and emits findings.

<=2025-11-25STDIO2 tools

42/100

Definition Quality47

Protocol Readiness47

Supportability30

Spec posture

Inferred effective spec <=2025-11-25·current baseline 2026-07-28

Static source inference · medium confidence · detected: stateful session

Deprecated protocol patterns detected

Summary

Two tools with complete schemas and descriptions, but both suffer from domain-specific naming that obscures intent for general LLM agents. 'classify_inventory' and 'caveats' are jargon-heavy and lack the verb_noun clarity pattern (e.g., 'analyze_mcp_security', 'generate_access_controls'). Descriptions are adequate (120-180 chars) but lack actionable context: no guidance on when to call each vs. the other, no examples of expected input structure, no explanation of output fields. Input schemas are present but minimal, only one required parameter (inventory object) with no type validation, format constraints, or nested field documentation. No output schema documented. Parameters lack descriptions of what 'inventory' should contain. Error handling not visible in tool definitions.

Tools (2)

caveatsread onlysource verified47/100

Turn an mcp-recon.inventory.v1 into an `mcp-recon/v0.1/caveats` artifact — capnagent-ready deny/scope plans with caveat-DSL predicates and rule provenance for every authority-relevant tool. The Find→Bind handoff.

classify_inventoryread onlysource verified47/100

Classify an mcp-recon.inventory.v1 document against deterministic, rule-based detectors (OWASP LLM Top 10 / NIST AI RMF / MITRE ATLAS). Same inventory in, same findings out — no LLM in the loop. Returns a findings.v1 array as JSON-encoded text.

Issues (5)

HIGHtool ↗

Tool names use domain jargon ('classify_inventory', 'caveats') instead of action verbs. LLMs cannot infer intent from names alone. Should be 'analyze_mcp_security' or 'generate_access_controls'.

classify_inventorycaveats

HIGHtool-description ↗

Input parameter 'inventory' lacks description. LLMs cannot determine what structure or fields this object must contain. No guidance on required vs. optional fields, nested schema, or format.

classify_inventorycaveats

HIGHtool ↗

No output schema documented. LLMs cannot plan downstream actions or extract specific fields from results. Descriptions mention 'findings.v1 array' and 'caveats artifact' but no formal schema provided.

classify_inventorycaveats

MEDIUMtool-description ↗

No distinction between tools in descriptions. Both accept 'inventory' and return JSON; LLMs cannot determine when to call classify_inventory vs. caveats without trial and error.

Recommendations

Rename tools to action-verb form: 'analyze_mcp_security' (for classify_inventory) and 'generate_access_controls' (for caveats). This matches the verb_noun baseline and helps LLMs infer intent from the name alone.
Add parameter description for 'inventory': 'An mcp-recon.inventory.v1 JSON object containing servers array (each with name, version, tools array) and tools array (each with name, description, inputSchema). See https://github.com/euanmcrosson-dotcom/mcp-recon#inventory-schema for full spec.'
Document output schemas explicitly in tool descriptions. E.g., 'Returns findings.v1 JSON: {findings: [{id, rule, severity, tool_name, description}], metadata: {timestamp, version}}'.
Expand tool descriptions to clarify use cases: 'classify_inventory: Detects security risks in MCP tool definitions using OWASP LLM Top 10 rules. Call this first to audit a server. Returns risk findings.' and 'caveats: Generates deny/scope policies for capnagent based on inventory. Call after classify_inventory to produce enforcement rules.'
Add error handling guidance: 'If inventory is invalid, returns error with field path and expected format. Ensure inventory.servers and inventory.tools are arrays with required fields: name (string), description (string), inputSchema (JSON Schema object).'

Spec posture evidence

Inferred effective spec: <=2025-11-25.

Relies on Stateful initialize / Mcp-Session-Id (removed; protocol is stateless) - make each request self-contained

Score history

Overall score trend

First recorded score · v2 rubric

42/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	F	42	<=2025-11-25	v2