Agent8 MCP Server for game development with prompt and tool support
Agent8 MCP Server implements a single tool (search_code_examples) with a well-structured input schema and comprehensive parameter descriptions. However, critical gaps in output schema documentation, error handling guidance, and composition patterns prevent a higher score. The tool has good naming (verb_noun pattern: search_code_examples) and detailed parameter descriptions, but lacks documented return structure and error recovery guidance. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) present despite the tool being marked READ_ONLY. Output is returned as raw JSON string in a text field, with no documented schema for the returned structure.
Searches and retrieves relevant game development code examples from a vector database based on specific game development requirements or programming challenges. This tool performs semantic search to find code snippets and examples that match the user's game development needs. It analyzes both the user message and associated tags to identify the most appropriate game code examples from the database.
Output schema not documented. Tool returns JSON stringified result in a text field, but the structure of the returned data (fields, types, pagination) is not specified. LLMs cannot plan downstream operations or extract relevant fields without documented output schema.
No error handling guidance or recovery instructions. Tool description lacks information about failure modes (e.g., vector DB unavailable, empty results, invalid tags). No actionable error messages to guide LLM behavior on failure.
Tool marked READ_ONLY in risk field but lacks toolAnnotations. Modern MCP spec (2026-07-28) supports readOnlyHint in tool annotations. Tool should declare readOnlyHint=true to formally communicate its safe nature to clients.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | C | 64 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 19 | - | v1 |
Result limit and pagination not addressed. Tags parameter accepts array of strings with no documented limit. Large tag arrays or result sets could exceed context windows. No mention of result count limits or pagination support.
Parameter relationship and constraints under-documented. Tags and userMessage descriptions are verbose but lack concrete constraints: minimum/maximum lengths, maximum number of tags, tag character restrictions, language enforcement ('ENGLISH ONLY') not validated server-side.