A personal finance management system with MCP tools for transaction querying, categorization, counterparty management, and Claude Agent SDK integration
Breadbox demonstrates solid tool design with comprehensive parameter schemas and clear naming conventions. All 7 tools follow verb_noun patterns (query_, list_, get_, update_, assign_, unlink_). Descriptions are present and substantive (100-250 chars typical), explaining WHAT the tool does and WHEN to use it. Input schemas are well-structured with type definitions and parameter descriptions. However, output schemas are not documented, LLMs cannot predict response structure for downstream chaining. Error handling guidance is absent; tools do not indicate what to do on failure. Some parameter descriptions could be more prescriptive about constraints (e.g., date format validation, enum values for sort_by/search_mode). The query_transactions tool is particularly strong with 15 well-described parameters and clear filtering semantics; counterparty tools show good composition (separate read/write operations). Missing: output field documentation, error recovery guidance, and explicit idempotency markers for write operations.
Resolve or create a counterparty and optionally link transactions to it, with optional enrichment fields
Retrieve a single counterparty by short ID or UUID
List all counterparties (merchants/payees) in the system
Query transactions with filtering by date range, account, category, amount, pending status, flagged status, search terms, tags, and pagination. Supports count_only mode for result sizing.
Aggregate transaction data by grouping (category, month, week, day, or category_month) with optional date range and account filtering
Detach transactions from a counterparty
Update counterparty metadata (name, website URL, logo URL, category, merchant category code)
Output schemas not documented. LLMs cannot predict response structure (field names, types, nested objects) for downstream tool chaining. query_transactions returns paginated results but response format is not specified.
Error handling guidance absent. Tools do not indicate what errors are possible, whether they are retryable, or what the LLM should do next. No recovery hints (e.g., 'If counterparty not found, try list_counterparties').
Enum constraints not formalized. sort_by accepts 'date|amount|provider_name' and search_mode accepts 'contains|words|fuzzy', but these are described as strings without explicit enum declarations. LLMs may hallucinate invalid values.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 55 | 2026-07-28+ | v2 |
Write operations lack idempotency markers. update_counterparty, assign_counterparty, and unlink_counterparty_transaction modify state but do not declare whether they are idempotent or require confirmation. Agents cannot safely retry on ambiguous failures.
Parameter constraints under-specified. Date parameters (start_date, end_date) state format 'YYYY-MM-DD' but do not clarify inclusive/exclusive boundaries or timezone handling. Numeric ranges (limit, amount) lack min/max bounds in schema.