anyapi-mcp has well-structured tool definitions with complete schemas and descriptions for all 8 tools. Naming follows verb_noun convention (add_api, list_apis, search_operations, execute, login, logout, remove, regenerate). Descriptions are substantive (100-200 chars typical). However, several tools lack output schema documentation, and error handling guidance is minimal. The execute tool is particularly high-risk (arbitrary code execution) but lacks confirmation/dry-run patterns. Parameter descriptions are generally good but some lack format/constraint details.

• Document output schemas for all tools. E.g., list_apis returns [{id, name, baseUrl, operationCount, authStatus}]; search_operations returns [{operationName, description, apiId, parameters}]. This enables LLMs to chain tools correctly.
• Add dry-run support to execute: accept a 'dry_run' boolean parameter that parses and type-checks the code without running it. Return validation errors so agents can iterate safely.
• Refactor add_api to accept token/app_password via server-side environment variables (ANYAPI_TOKEN_<api_id>) or OS keychain lookup, not as parameters. Document this in the description.
• Enhance error messages with recovery hints. E.g., 'API with id "petstore" already exists. Use force=true to overwrite, or choose a different id.' instead of a bare error.
• Add a 'limit' parameter to list_apis (default 50) and document pagination. If 100+ APIs are registered, returning all at once wastes tokens and risks context overflow.