AgentValet MCP server — lets AI agents call approved platforms via the AgentValet proxy
AgentValet MCP server has 8 tools with basic descriptions and parameter schemas, but significant gaps limit production readiness. All tools have descriptions (10-100 chars, mostly adequate), and most parameters include type and description. However, output schemas are not documented, error handling lacks recovery guidance, and several tools combine multiple concerns. The server targets a specialized domain (credential brokering + platform access) but lacks the polish expected for A-grade tooling. Naming is generally clear (verb_noun pattern), but composition issues and missing output documentation are material weaknesses.
Self-register an AI agent with AgentValet
Check the registration status of an agent
Evaluate authorization decision for a platform and scope using Authzen
List pending approval actions for this agent
List approved platforms and their granted scopes for this agent
Report a diagnostic message (error, warning, info) to AgentValet
Request access to a platform that is not yet granted
Output schemas not documented. No tool returns a documented response structure. LLMs cannot plan downstream calls or extract required fields (e.g., what does list_platforms return? Is it {platforms: [{id, scopes}]} or {data: [...]}?). This forces agents to guess and wastes tokens on trial-and-error parsing.
use_platform combines multiple concerns: credential injection, scope validation, HTTP method dispatch, and approval flow. Should split into separate tools (call_platform_endpoint, request_platform_approval) so agents can compose them independently.
use_platform has both 'body' and 'data' parameters (legacy alias). Ambiguous which to use; LLMs may pass both or neither. Consolidate to single 'body' parameter and document in description.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 61 | 2026-07-28+ | v2 |
Call an approved platform endpoint with optional credential injection and approval flow
Error handling lacks recovery guidance. No tool description explains what to do on failure (e.g., 'If scope denied, call request_platform_access()'). Agents cannot self-correct without explicit error classification and next-step hints.
agent_register and agent_status form a two-step flow but lack clear documentation of the workflow. Description should explain: 'Call agent_register first to get a token, then poll agent_status with that token until status is approved.'