Azure Chat Assistant MCP Server — standalone chat assistant powered by Azure AI models with support for multiple LLM providers (Azure, OpenAI, Anthropic, Google, DigitalOcean, AWS Bedrock, Puter)
Server has 5 tools with basic descriptions and input schemas, but critical gaps undermine quality. Tool names lack action verbs (chat, configure, reset, clear_cache, models). Descriptions are present but generic (10-50 chars for most). Input schemas exist but lack proper type definitions and parameter descriptions. The 'configure' tool is dangerously vague, no enum for 'setting' parameter, no validation guidance. Error handling is absent from visible code. No output schemas documented. Composition is poor: 'configure' conflates viewing and changing settings; 'reset' and 'clear_cache' are similar concerns. No evidence of idempotency, pagination, or recovery guidance.
Send a message to the LLM, get a response (with conversation history)
Clear the in-memory response cache
View/change settings dynamically (model, API key, region, etc.)
List available models and test connectivity
Clear conversation history
Tool names lack action verbs. 'chat', 'configure', 'reset', 'clear_cache', 'models' do not start with clear action verbs (send_, get_, list_, update_, delete_). LLMs cannot infer intent from these names alone.
'configure' tool conflates two responsibilities: viewing settings (read) and changing settings (write). Should split into get_config and update_config. Current design forces LLM to guess whether to pass a value.
'configure' setting parameter has no enum, validation rules, or description of valid values. LLM can pass arbitrary strings. No guidance on which settings exist or what values they accept.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields. 'chat' response structure unknown; 'models' return format undocumented.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 40 | 2026-07-28+ | v2 |
No error handling guidance visible. No recovery hints, retryability classification, or actionable error messages. If 'chat' fails, LLM has no guidance on what to do next.