Official SDK and platform for Melaya Agent Builder, Mobile Device Control, and Trading. Includes market-data streaming, agentic pipelines, and trading engine across 70+ venues.
This is a benchmarking framework, not a production MCP server. Tool definitions are scattered across Python benchmark files and C# SDK code, making them difficult to verify as actual MCP tool registrations. Most tools lack proper input schemas, parameters are described in docstrings but not in structured JSON Schema format. Descriptions exist but are benchmark-focused rather than LLM-optimized. No evidence of error handling patterns, output schemas, or security considerations. The 'call' tool accepts generic dicts without type constraints. Tool names like 'dispatch' and 'async_tool' are vague and don't follow verb_noun conventions.
Get the current server version string. Public endpoint — no auth required.
Zero-cost mock tool: returns immediately with a fixed response. Used by the dispatch bench so we measure ONLY runner overhead.
One LLM turn through the wrapper. The hot path the bench measures. Packs message history into provider format, injects tools spec, awaits provider, unpacks response and records cost.
Steady-state dispatch. Mirrors the runtime's toolkit hot path: 1. Name lookup 2. kwargs merge 3. Await the tool func 4. Wrap into a ToolResponse if the func didn't already
Register a tool. Mirrors `Toolkit.register_tool_function`'s hot-path behaviour: the func is stored under its name (or the override), preset kwargs are merged in on every call.
Sync variant of `async_tool` — same zero-cost return, no executor offload modelled. Used by the dispatch bench to isolate the sync-vs-async dispatch delta.
No visible JSON Schema input definitions for any tool. Parameters described in docstrings only (e.g., 'tool_func' as callable, 'messages' as list[dict]) but not in structured schema format.
Tool names violate verb_noun convention. 'dispatch', 'async_tool', 'sync_tool' are vague and don't clearly signal action. 'VersionAsync' mixes camelCase with unclear intent. Should be 'get_version', 'call_tool', 'invoke_async_tool'.
Output schemas not documented. Tools like 'call' and 'dispatch' return responses but no structured schema is visible describing return fields, types, or pagination. LLMs cannot plan downstream calls without knowing output structure.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 33 | 2026-07-28+ | v2 |
No error handling guidance. Tools lack descriptions of failure modes, recovery steps, or error classification (retryable vs. fatal). 'dispatch' and 'call' could fail in many ways but provide no actionable error messages.
Generic parameter types. 'call' accepts 'tools' as list[dict] and 'messages' as list[dict] with no field-level schema. 'register_tool_function' accepts 'preset_kwargs' as dict with no constraints. This invites hallucinated values and makes validation impossible.