Loading ToolBench
Local FastAPI chat backend with Azure Foundry Local integration. Provides OpenAI-compatible chat endpoints that connect to Azure Foundry Local running on Windows from WSL using port forwarding.
This is NOT an MCP server. It is a FastAPI HTTP service that exposes two chat endpoints (/chat, /chat/stream). The tools are inferred from HTTP route documentation, not registered via MCP protocol. No MCP tool definitions, schemas, or protocol implementation are visible in the source. The descriptions are verbose marketing copy (200+ chars) rather than LLM-optimized parameter documentation. Input schemas are partially visible but lack proper JSON Schema type definitions for nested objects. No error handling guidance, no output schemas documented, no tool annotations. This fails the fundamental requirement: it is not an MCP server at all.
Single chat completion endpoint - PRODUCTION READY. This endpoint provides complete responses from Azure Foundry Local. Key features: Max 2000 tokens for long responses (configurable), 120 second timeout for complex processing, Temperature 0.7 for balanced creativity/consistency, Filters reasoning tokens if present, Returns complete response at once (no streaming). Perfect for: Lists (like "20 facts about embeddings"), Long explanations, Code generation, Complex analysis
Streaming chat completion endpoint - EXPERIMENTAL. Returns real-time stream of tokens as they're generated by the model. Uses Server-Sent Events format with NDJSON data. Features: Real-time token streaming for immediate feedback, Error handling with graceful degradation, Timeout protection (60 seconds), Connection interruption detection. Best for: Interactive conversations, Real-time user experience, When immediate feedback is needed. Note: May occasionally experience connection issues or timeouts with Azure Foundry Local. Use /chat for guaranteed complete responses.
Not an MCP server. This is a FastAPI HTTP service with no MCP protocol implementation. No tool registration via MCP, no capabilities negotiation, no MCP transport layer. Cannot be used as an MCP server by any MCP client.
Tool definitions are inferred from HTTP routes, not explicitly registered. No MCP tool schema objects visible.
Input schema for both tools lacks proper JSON Schema type definitions. 'messages' is array of objects, but nested 'role' and 'content' properties have no 'type' field.
Descriptions are verbose marketing copy (200+ chars) focused on features, not LLM-optimized parameter guidance. No parameter descriptions for 'messages', 'role', or 'content'.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 26 | 2026-07-28+ | v2 |
No output schema documented. LLMs cannot plan downstream calls or extract fields. No error handling guidance (e.g., timeout recovery, Azure Foundry connection failures). No tool annotations (readOnlyHint, destructiveHint).