This server has severe definition quality issues. While all 18 tools have names and descriptions, the descriptions are extremely generic (mostly 1-2 sentences with no actionable detail), parameter descriptions are minimal or absent, output schemas are completely undocumented, and error handling is not visible. The tool names follow a wrapper pattern rather than action verbs (nuclei_scan_wrapper vs. scan_nuclei). Many parameters lack type information in descriptions and have no guidance on valid ranges, formats, or dependencies. No evidence of schema documentation for tool outputs. The server is primarily a thin wrapper around security CLI tools without the abstraction and structure expected of production-grade MCP servers.
Tool names use generic '_wrapper' suffix instead of action verbs. LLMs cannot infer intent from 'nuclei_scan_wrapper' vs. 'scan_nuclei_target'. Names should start with action verbs (scan_, enumerate_, discover_, probe_) to enable LLM self-service without reading full descriptions.
Descriptions are extremely generic and lack actionable detail. Examples: 'Wrapper for running a Nuclei security scan' (43 chars), 'Wrapper for running ffuf fuzzing' (33 chars). No explanation of WHEN to use each tool, WHAT makes it distinct from similar tools, or WHAT the output contains. LLMs cannot disambiguate between nuclei, ffuf, wfuzz, or other scanning tools from these descriptions.
Rename all tools from '<action>_wrapper' to '<action>_<target>' pattern. Examples: 'nuclei_scan_wrapper' → 'scan_nuclei_templates', 'ffuf_wrapper' → 'fuzz_web_endpoints', 'nmap_wrapper' → 'scan_network_ports'. This enables LLMs to infer intent from names alone.
Expand descriptions to 100-200 characters with explicit WHEN/WHY guidance. Example for nuclei: 'Scan a target with Nuclei templates to detect known vulnerabilities. Use this for template-based security scanning; use nmap_wrapper for port/service discovery instead. Returns JSON with vulnerability findings.'
Document ALL output schemas explicitly. Create a separate 'Output' field in each tool definition. Example: 'Output: JSON object with keys: {"vulnerabilities": [{"template": string, "severity": string, "matched": boolean}], "scan_status": string, "target": string}'. This enables agents to chain tools and extract relevant fields.
Add JSON Schema enums for all constrained parameters. Example: 'severity' should have enum: ["critical", "high", "medium", "low", "info"]. 'scan_type' in nmap should enumerate valid Nmap scan types. 'output_format' should enumerate ["json", "text"].
Add parameter descriptions that explain format, range, and purpose. Example for 'wordlist': 'Path to wordlist file (required; must be readable by the container; typically /wordlists/* or custom path; no path traversal allowed). If file not found, the scan will fail with a clear error message.'
Implement strict input validation with actionable error messages. When a tool receives an invalid value, return a structured error like: {'status': 'error', 'code': 'INVALID_SEVERITY', 'message': 'Invalid severity: got "super-critical", must be one of: critical, high, medium, low, info', 'valid_values': [...]}. Do NOT return raw stack traces.
Score history
Overall score trend
↓ 8 points across a rubric change (v1 → v2)
45/100
Scored
Grade
Overall
Spec posture
Rubric
2026-09-22
F
45
2026-07-28+
v2
2026-03-09
D
53
-
v1
read onlysource verified43/100
Wrapper for running Gospider web crawling.
hashcat_wrapperread onlysource verified33/100
Wrapper for running Hashcat password cracking.
httpx_wrapperread onlysource verified37/100
Wrapper for running HTTPX scan.
ipinfo_wrapperread onlysource verified37/100
Wrapper for getting IP information using ipinfo.io.
nmap_wrapperread onlysource verified35/100
Wrapper for running Nmap scan.
nuclei_scan_wrapperread onlysource verified40/100
Wrapper for running a Nuclei security scan.
sqlmap_wrapperread onlysource verified37/100
Wrapper for running SQLMap scan.
subfinder_wrapperread onlysource verified37/100
Wrapper for running Subfinder subdomain enumeration.
Output schemas are completely undocumented. All tools return a string, but what is the structure of that string? JSON, plaintext, or raw tool output? Are there pagination fields? Failure indicators? Response fields that downstream tools need (e.g., target_id, scan_id for chaining)? An agent cannot plan multi-step workflows without knowing what each tool returns.
Parameter descriptions are minimal or absent. Example: 'filter_code' is described as 'HTTP status code to filter out' but no explanation of format (e.g., '404' as string or 404 as int), no guidance on what happens if this is omitted, no mention of whether multiple codes are supported. Similarly, 'wordlist' is just 'Path to wordlist file' with no guidance on format, location, or what happens if the file doesn't exist.
Security concern: Tools accept file paths and wordlist paths as user input with no validation visible in the source. A malicious agent input like '../../../etc/passwd' or '/dev/zero' could cause path traversal, resource exhaustion, or data exfiltration. All file path parameters must be sanitized, and only whitelisted directories should be accessible.
No error handling visible. Tools return strings, but how are errors communicated? If nuclei fails to reach a target, does the string contain 'error:', a JSON error object, or raw stderr? LLMs cannot distinguish between success and failure without explicit error structure and recovery guidance.
No enums for constrained parameters. 'severity' accepts 'critical, high, medium, low, info' but this is documented in a free-form description, not as a JSON Schema enum. 'output_format' accepts 'json' or 'text' but again, no enum constraint. LLMs often hallucinate invalid values like 'high-critical' or 'xml' because they rely on examples in text rather than formal constraints.
Parameter dependencies are undocumented. Example: 'hashcat_wrapper' accepts 'hash_type' (0 for MD5, 1000 for NTLM) but the valid values are numeric codes from the Hashcat docs, not enums. An LLM has no way to know valid values and will hallucinate. Similarly, 'scan_type' in nmap_wrapper defaults to 'sV' but valid types (sS, sT, sU, sA, sW, sM, sN, sF, sX, sP, etc.) are not enumerated.
No idempotency or retry safety documented. Tools like 'send_email' or 'create_ticket' would need explicit idempotency markers. For security scanning tools, it's less critical, but tools that interact with state (if any downstream) should declare whether repeated calls with the same inputs are safe. No toolAnnotations (readOnlyHint, destructiveHint, idempotentHint) are visible.
Add security guards for file path parameters. Validate that all file paths are within a whitelisted directory (e.g., /wordlists, /data), deny absolute paths outside that scope, and reject path traversal patterns like '../'. Log all file access attempts for audit trails.
Implement timeout handling for all external tool calls. Set explicit timeouts (e.g., 30s for network scans, 60s for intensive operations), catch timeout exceptions, and return a structured error with retry guidance: {'status': 'timeout', 'message': 'Scan exceeded 30s timeout. The target may be slow or unreachable. Retry with reduced scope or longer timeout if appropriate.'}
Batch similar operations where agents would call them in loops. Add 'scan_multiple_nuclei_targets' that accepts an array of targets instead of requiring 18 separate calls to nuclei_scan_wrapper. Reduce per-tool latency and token overhead.
Document pagination and result limits. If any tool returns large result sets, specify: 'Results are capped at 1000 items. If you need more, contact support or use the tool multiple times with refined filters.' This prevents context explosion.
Add tool annotations (readOnlyHint, idempotentHint) to the FastMCP decorator. All tools here appear to be read-only (safe to call repeatedly), so mark them: @mcp.tool(readOnlyHint=True, idempotentHint=True). This helps agents optimize for retries and parallel execution.
Create a shared data model for common outputs. Define reusable schemas for scan results, enumeration results, and errors so agents learn patterns and can chain tools confidently. Example: all enumeration tools should return {"results": [{"target": string, "value": string}], "count": int, "total_time_seconds": float}.