Static source inference · medium confidence · detected: Logging
Deprecated protocol patterns detected
Summary
This MCP server has 18 tools across Internet registries (RIPE, ARIN, APNIC, AFRINIC, LACNIC). While tool names follow verb_noun conventions consistently, the implementation suffers from critical gaps in schema documentation and parameter descriptions. Only 2 tools (ripe_validate_route_object, arin_validate_route_object) have visible input schemas in the source; for the remaining 16 tools, schemas are registered but not shown in the provided code. Descriptions vary widely, some tools have substantive descriptions (route validators at ~120 chars), but most registered tools in register.ts lack visible documentation. Parameter descriptions are present for the two route validator tools but missing entirely for the 16 other tools. No output schemas are documented. Error handling exists in the route validators (validation errors, service disabled, HTTP errors) but is absent from or unclear in most other tools. The server implements a TTL cache and basic HTTP error recovery, showing some production intent, but lacks comprehensive error guidance and structured output documentation that LLMs need for reliable tool selection and chaining.
16 of 18 tools lack visible input schemas in source code. Tool registration in register.ts shows only tool names; detailed schema definitions are not shown, making it impossible to verify parameter types, constraints, or descriptions. Per hard scoring rule: if tool definition is not visible in source, cap overall at 50.
Missing or minimal tool descriptions. 16 of 18 tools have no visible description in the provided source code. Only ripe_validate_route_object and arin_validate_route_object have substantive descriptions (120 chars each). Descriptions are LLM's primary mechanism for tool selection.
Recommendations
Add visible input schemas to all 16 undocumented tools in src/register.ts. Each tool registration must include full Zod/JSON Schema definitions with type, description, and constraints for every parameter. Example: for whois_query, define parameters like 'query_string' (string, required, 1-255 chars, 'IP, ASN, domain name, or email to search in WHOIS'), 'query_type' (enum: ip|asn|domain|email|name), and 'rir' (enum: ripe|arin|apnic|afrinic|lacnic).
Write substantive descriptions for all 16 undocumented tools. Each must be 50 - 200 characters, answering: What does it do? When should LLM call it? What does it return? Example: 'ripe_whois_query: Search the RIPE NCC WHOIS database for IP addresses, AS numbers, or network objects. Returns registration details, contact info, and routing data. Use this to audit network ownership or validate registration status.' Current 'whois_query' conveys almost nothing.
Document output schemas for all tools. Create a types.ts or schema file declaring the return structure for each tool. Example for ripe_whois_query: return {ok: boolean, data?: {matches: [{object_type: string, handle: string, created: ISO8601, last_modified: ISO8601}], total_count: number, query_type: string}, error?: {code: string, detail: string}}. This enables LLMs to plan downstream chaining.
Expand error handling to all 16 undocumented tools. Implement consistent error categorization: invalid_input (malformed query), not_found (no results), service_unavailable (RIR down), rate_limit (quota), permission_denied (auth required). Return actionable messages: 'Invalid ASN: expected format ASxxxxx or decimal number 1-4294967295. Got: "AS65000x". Did you mean AS65000?'
Spec posture evidence
Inferred effective spec: <=2025-11-25.
Relies on Logging (deprecated) - log to stderr or use OpenTelemetry
No documented output schemas. None of the 18 tools have documented return types or output field descriptions visible in source. LLMs cannot plan downstream tool calls or extract result data when output structure is undocumented. Required baseline: 100% of A+ tools have documented return types.
Inconsistent error handling across tools. The two route validator tools (ripe_validate_route_object, arin_validate_route_object) have structured error responses with categorized error codes (invalid_prefix, service_disabled, http_error, validation_error) and actionable detail messages. However, 16 other tools provide no visible error handling logic in the source, making it impossible for LLMs to know how to recover from failures.
Parameter descriptions missing for 16 tools. Only ripe_validate_route_object and arin_validate_route_object have visible parameter annotations. Both route validators describe 'prefix' (IP prefix, with validation note) and 'origin_asn' (AS number, note about no 'AS' prefix).
Tool naming could be more specific for disambiguation. Tools like 'ripe_whois_query', 'arin_whois_query', etc. are generic action verbs + resource. However, distinct query/search operations should clarify their intent. Consider: does whois_query search by IP, ASN, name? Is it a lookup or search? Generic names like 'query' force LLMs to guess when multiple similar tools exist across five RIRs (25+ similar tool names). Suggest names like 'search_whois_by_ip', 'search_whois_by_asn', or 'lookup_asn_contact'.
Rename generic query tools for clarity. Replace 'ripe_whois_query' with 'ripe_search_by_ip', 'ripe_search_by_asn', 'ripe_search_by_domain' (separate tools), OR add an explicit 'query_type' parameter with enum constraint and describe what each type returns. This prevents LLM confusion across 25+ similar tool names.
Add pagination support. If any query tool can return large result sets (common for WHOIS), add 'limit' (1 - 100, default 20) and 'offset' or 'cursor' parameters. Return 'total_count' and 'has_more' in response so LLM can iterate if needed. Per pattern:paginated-result, unbounded results blow context windows.
Document RIR-specific differences in tool descriptions. RIPE, ARIN, APNIC, AFRINIC, LACNIC handle queries differently (some require auth, some have rate limits, result formats vary). Add 'Note: ARIN queries require authentication. Unauthenticated queries return limited data.' This helps agents choose the right tool and set expectations.
Create a shared 'whois_query' tool that accepts a 'rir' parameter (enum: all|ripe|arin|apnic|afrinic|lacnic) instead of 18 RIR-specific tools. Reduces tool explosion, clarifies intent, and lets LLM specify which registry to query. Alternatively, if per-RIR tools are intentional, use consistent naming: 'query_ripe_whois', 'query_arin_whois' (verb_rir_noun pattern).
Add validation and constraints to route validator parameters. prefix: enforce CIDR notation (regex '^([0-9]{1,3}\.){3}[0-9]{1,3}/[0-9]{1,2}$|^([0-9a-fA-F]{0,4}:){2,7}[0-9a-fA-F]{0,4}/[0-9]{1,3}$'), origin_asn: enforce range 1 - 4294967295 and integer type. Return clear validation errors: 'Invalid prefix: "192.0.2.256/24", octet 256 exceeds 255. Valid IPv4 prefixes are 0.0.0.0/0 to 255.255.255.255/32 in CIDR notation.'
Implement caching headers in HTTP transport. The route validators use TTLCache (300s). Document cache behavior in tool descriptions so agents know when results may be stale. Example: 'Results cached for 5 minutes. For real-time validation, call with cache_bust=true parameter.' This prevents agents from assuming fresh data.
Add tool annotations for structured LLM reasoning. Set readOnlyHint=true on all 18 tools (all are READ_ONLY per risk assessment). Optional: add idempotentHint=true on search/lookup tools so LLMs know retries are safe. Per spec 2026-07-28, tool annotations enable more intelligent agent planning.