A lightweight MCP proxy and sidecar that sits between MCP clients and upstream servers, providing audit logging, PII masking, request replay, and a web console for monitoring and management.
MCP Arc exposes 7 tools across two example servers. (3) No parameter descriptions anywhere, LLMs cannot infer what email, national_id, card_number, cvv, amount, api_key, or api_key mean in context. (4) No output schemas documented. (5) No error handling guidance. (6) Sensitive parameters (cvv, api_key, card_number) exposed as tool inputs, violating secret-injection pattern. Tools 1-4 have better structure but still lack parameter descriptions and output documentation.
Charge a credit card.
charge_card
Echo back the arguments (demo tool for MCP Arc)
lookup_customer
Look up a customer by email / national ID.
send_email
Send an email via a provider API.
Tools 5-7 have empty input schemas (no properties defined). These are duplicates of tools 2-4 with degraded definitions.
Tools 5-7 have trivial descriptions (tool name only, <20 chars). Descriptions must explain WHAT the tool does, WHEN to use it, and any prerequisites.
No parameter descriptions anywhere. LLMs cannot infer meaning of 'email', 'national_id', 'card_number', 'cvv', 'amount', 'api_key' from names alone. Each parameter requires a description explaining format, constraints, and purpose.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 40 | 2026-07-28+ | v2 |
Sensitive parameters (cvv, api_key, card_number) exposed as tool inputs. Violates secret-injection pattern, agent traces log all parameters, leaking secrets into logs and prompt history. Use server-side secret injection via environment variables.
No output schemas documented. LLMs cannot plan downstream tool calls or extract required fields without knowing response structure. Document what fields each tool returns and their types.