Personal finance MCP server for managing transactions, categories, and user authentication
FinanceManager has 9 tools with basic naming and descriptions, but critical gaps in schema completeness, parameter documentation, and error handling. All tools have descriptions (10-50 chars, below the 50-200 char optimum), but most parameters lack type constraints and validation guidance. No output schemas documented. Error responses are generic strings rather than actionable recovery guidance. Global AUTH_TOKEN state management violates stateless design. No tool annotations (readOnlyHint, destructiveHint). Composition is reasonable (single responsibility per tool), but parameter descriptions are minimal.
Adds a new transaction.
Deletes a transaction.
Deletes a user.
Fetch all categories from the personal finance service.
Fetch all transactions from the personal finance service.
Logs in a user and stores the authentication token.
Logs out the current user.
Global AUTH_TOKEN state violates stateless request handling. MCP protocol requires each request to be self-contained; storing auth in module-level variables breaks isolation and prevents concurrent agent sessions.
No output schemas documented. Tools return plain strings (e.g. 'Registration successful') or markdown tables, but LLMs cannot parse unstructured text reliably. Downstream tools cannot extract IDs or references for chaining.
Error responses are generic strings ('Registration failed: {res.text}', 'You must be logged in...') without recovery guidance. LLMs cannot determine if errors are retryable, user-fixable, or fatal.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 49 | 2026-07-28+ | v2 |
Registers a new user and logs them in.
Updates an existing transaction.
Parameter descriptions are minimal (1-5 words). 'Date of the transaction' lacks format guidance (ISO 8601? MM/DD/YYYY?). 'Amount of the transaction' lacks range constraints. LLMs cannot validate inputs without explicit constraints.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Agents cannot distinguish safe reads from destructive writes without explicit hints. delete_transaction and delete_user should be marked destructive.