A policy-enforcing gateway MCP server that sits between clients and downstream MCP servers, filtering tools and resources based on active profiles with human-in-the-loop approval for profile switches
Aegis-MCP defines 2 tools with minimal documentation. Both tools have descriptions (10-20 chars each), but lack parameter descriptions entirely. Input schemas are present but sparse: only 'name' and 'id' string parameters with no type constraints, enums, or validation guidance. No output schema is documented. Error handling is minimal, tools return plain text ('code=...', 'approved', 'pending') without structured error classification or recovery guidance. The tool names use verb_noun style (set_profile, approval_status) but 'approval_status' is ambiguous, does it check status or apply approval? The description clarifies, but the name alone is unclear. No security annotations, no idempotent hints, no permission gates documented. Overall, definitions are skeletal and would not pass production code review.
Apply a pending profile-switch approval by id, or report it is still pending.
Request a switch to a different Aegis profile. Some switches require human approval.
Parameter descriptions missing entirely. 'name' and 'id' parameters have no descriptions explaining what values are valid, expected format, or constraints.
No output schema documented. Tools return plain text responses ('code=...', 'approved', 'pending') with no structured schema, field definitions, or type information for LLM parsing.
Tool name 'approval_status' is ambiguous. Description says it both checks AND applies approval, but the name suggests read-only status check. Rename to 'apply_approval' or split into separate tools.
No error handling guidance. Plain text responses ('code=...') do not indicate success/failure, retryability, or next steps. LLM cannot distinguish between valid and error states.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 41 | 2026-07-28+ | v2 |
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Both tools modify state (profile switches, approval application) but lack destructiveHint or idempotent markers.