MCP Server for Brex API Integration - provides access to Brex accounts, transactions, expenses, budgets, and spend limits
The server registers 21 tools with visible schemas and descriptions in src/tools/index.ts. However, systematic quality issues prevent a higher score: (1) Descriptions are minimal (10-80 chars) and lack actionable context about WHEN to use tools or their side effects. Most lack dependency hints, prerequisites, or explanation of what distinguishes them from similar tools. (2) Parameter descriptions are sparse, many params like 'expand', 'status', 'limit' lack context about their valid ranges, formats, or constraints. (3) No output schemas are documented; LLMs cannot plan downstream operations. (4) Write operations (upload_receipt, match_receipt, update_expense) lack confirmation patterns or dry-run support despite modifying state. (5) Error handling is not visible in the tool definitions, no guidance on recovery or categorization. (6) Tool composition shows duplication and unclear boundaries: get_budget vs get_budgets, get_spend_limit vs get_spend_limits, and redundant listing tools (get_all_accounts, get_accounts via account_id parameter) suggest either incomplete implementation or missing consolidation. The tooling is functional for simple read operations but lacks production polish.
Get details for a specific Brex account
LIST: Get all Brex accounts with pagination support
LIST: Get all Brex card expenses with pagination support
LIST: Get all Brex expenses with pagination support
Get a budget by ID (read-only). Returns the complete budget object. Example: {"budget_id":"budget_123"}
Get a budget program by ID (read-only). Example: {"id":"bp_123"}
Descriptions lack actionable context. Many descriptions (e.g., 'Get a budget by ID (read-only)', 'Get expenses for a Brex account') are under 50 characters and do not explain WHEN to use this tool vs a similar one, what it returns, or any prerequisites.
Parameter descriptions are minimal or absent. Parameters like 'expand', 'cursor', 'limit', 'status', and 'updates' lack descriptions explaining their valid ranges, formats, or constraints. The 'updates' parameter in update_expense is described only as 'Fields to update', LLMs cannot infer which fields are valid or what format they expect.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 55 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 29 | - | v1 |
List budget programs (read-only). Returns complete budget program objects. Example: {"limit":10,"budget_program_status":"ACTIVE"}
List budgets (read-only). Example: {"limit":10}
Get a single card expense by ID. Returns the complete card expense object. Example: {"expense_id":"expense_123","expand":["merchant"]}
Get complete statements for the primary card account. Returns full statement objects. Example: {"limit":10}
LIST: Primary card transactions. Returns complete transaction objects. Example: {"limit":10,"posted_at_start":"2025-08-01T00:00:00Z","expand":["merchant"]}
Get cash account statements by account ID. Returns complete statement objects. Example: {"account_id":"cash_acc_123","limit":5}
LIST: Cash transactions (requires cash scopes). Example: {"account_id":"cash_acc_123","limit":10}. Returns complete transaction objects.
Get a single expense by ID. Returns the complete expense object. Example: {"expense_id":"expense_123","expand":["merchant"]}
Get expenses for a Brex account
Get a spend limit by ID (read-only). Example: {"id":"sl_123"}
List spend limits (read-only). Example: {"limit":10,"status":"ACTIVE"}
Get transactions for a Brex account
Match a receipt to an expense
Update an expense
Upload a receipt for an expense
No output schemas documented. Tool definitions show input schemas but do not specify what fields are returned. LLMs cannot plan downstream tool calls or extract IDs needed for chaining (e.g., does get_budgets return budget_id? Is it 'id' or 'budgetId'?). Baseline: 100% of A+ tools have documented return types.
Write operations lack confirmation or dry-run support. upload_receipt, match_receipt, and update_expense are state-changing tools but have no description indicating they are destructive, no idempotent hint, and no mechanism for confirmation before execution. Agents may apply these inadvertently.
Tool naming shows redundancy and unclear composition. get_all_accounts, get_all_expenses, get_all_card_expenses appear to duplicate get_account, get_expenses, and get_card_expenses with pagination. Unclear whether get_transactions (accepting accountId) is the same as get_card_transactions or get_cash_transactions. Similar tool names cause LLMs to conflate them.
Pagination parameters lack constraints. 'limit' appears in many tools without documented min/max (rubric example: 1 - 100). 'cursor' lacks explanation of format or origin. LLMs may pass invalid cursor values or request excessive limits.
No error handling guidance visible in tool definitions. No indication of what errors are retryable, user-fixable, or fatal. LLMs have no recovery hints if a lookup fails or an invalid parameter is passed. Baseline: error responses must tell the LLM what to do next.
Enum values in descriptions should be replaced with schema enum constraints. Several tools mention 'enum' values in descriptions (e.g., 'status' with 'ACTIVE|ARCHIVED'), these should be declared as enum in the schema, not in text, to be machine-parseable and prevent LLMs from inventing invalid values.