One command to turn any software into an MCP server
mcp-anything is a code-generation framework that produces MCP servers from API specifications. The 8 tools examined are from test fixtures (fake_socket_xmlrpc_app and fake_ws_protocol_app), not from mcp-anything's own runtime. Analysis reveals: (1) Tool naming is clear and verb-forward (add_item, get_item, list_items, etc.), following verb_noun convention. (2) Descriptions are present but terse (7-9 words each), falling below the 10-character minimum and averaging 60 chars vs. the 194-char production baseline. (3) Input schemas ARE present and typed (string, integer) with parameter descriptions, but lack depth. (4) No output schemas are documented, critical gap for agent chaining. (5) No error handling guidance, no idempotence hints for write operations, no recovery instructions. (6) The framework itself shows sophisticated tool-design rules in tool_design.py (CRUD grouping, arg_mapping for HTTP, operation discriminators), but these rules are not consistently applied to the test fixtures. The fixtures appear to be minimal stubs for testing the code generator, not production-quality tool definitions. This is a framework evaluation issue: the framework CAN generate good tools if given rich input, but the test examples are too thin.
Add an item to the inventory.
Get information about a registered device.
Get details of an inventory item by ID.
List all registered devices.
List all inventory items.
Register a new IoT device with the server.
Remove an item from the inventory.
Descriptions are terse (7-9 words, ~60 chars avg.) vs. production baseline of 194 chars. Missing context for LLM tool selection. Examples: 'Add an item to the inventory.' lacks WHEN to use it, what it returns, or prerequisites.
No output schemas documented for any tool. LLMs cannot infer response structure, breaking tool chaining. If add_item returns an item_id, and update_quantity accepts item_id, that linkage must be visible in the schema.
No error handling guidance. Destructive tools (remove_item) lack recovery hints. Write operations lack idempotence or dry-run support. LLMs have no guidance on retry eligibility or how to compensate on failure.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 55 | 2026-07-28+ | v2 |
Update the quantity of an existing inventory item.
list_items and list_devices lack pagination info (limit, offset, total_count). No guidance on result capping. Returning unlimited results risks context window overflow.
Parameter descriptions exist but are minimal (5-15 chars). No format constraints, range info, or examples. E.g., 'Quantity to add' lacks min/max bounds or clarification on negative values.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Agents cannot determine which operations are safe to retry or have side effects without explicit hints.