A collection of MCP servers for financial intelligence and banking data access, including Calendar, Gmail, Graph (Neo4j), Plaid, and Vision tools for AI agents to reason about finances
Foresight MCP servers exhibit significant definition quality gaps. While all 20 tools have names and descriptions, most descriptions are under 100 characters and lack actionable context for LLM selection. Input schemas are present but incomplete: many parameters lack descriptions (e.g., 'user_id' in get_spending_patterns has no description), and critical parameters like 'image_base64' in vision tools lack format guidance. No output schemas are documented. Error handling is absent from visible code. The calendar_mcp sample shows basic tool registration but no recovery guidance or error classification. Composition is reasonable (tools are single-purpose), but parameter naming inconsistencies (user_id vs no description) and missing field-level documentation reduce clarity. No tool annotations (readOnlyHint, destructiveHint) are visible despite clear read/write semantics.
Add a calendar reminder for an upcoming financial event like a bill due date or subscription renewal
Analyze a photo or PDF of a bill — utility, medical, insurance — and extract key information and flag anomalies
Find emails notifying the user of price increases on services they use
Create a financial alert in the graph and mark it as pending
Extract financial information from any document — lease, insurance policy, contract
Find all subscription-related emails to identify what services the user is paying for
Missing parameter descriptions across all tools. Parameters like 'user_id', 'image_base64', 'email_id' lack descriptions explaining what they control or their expected format. LLMs cannot infer meaning from names alone.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields without knowing response structure. Vision tools (scan_receipt, analyze_bill) especially need documented output fields.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear semantic differences. Tools marked WRITE (add_financial_reminder, create_alert_node, update_goal_progress) should declare destructiveHint=true to guide agent retry logic.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 47 | <=2025-11-25 | v2 |
Find all detected subscription patterns for a user from the graph
Detect transactions that are unusual compared to the user's normal spending patterns
Get current account balances for all accounts linked by the user
Get historical income and expense data for cashflow forecasting
Get the full content of a specific email by ID
Infer the user's payday schedule from their calendar and past patterns
Identify recurring charges and subscriptions the user is paying for
Get total spending grouped by category for a given time period
Get the user's historical spending patterns from the knowledge graph — top merchants, categories, trends
Fetch recent bank transactions for a user from their linked bank account
Get calendar events in the next N days that are financially relevant — bills, payday, rent, travel
Scan inbox for financial emails — bills, renewals, receipts, price changes — from the last N days
Extract structured data from a photo of a receipt — merchant, items, total, date, payment method
Update the current progress on a savings goal
Vision tool parameters lack format constraints. 'image_base64' parameter in scan_receipt, analyze_bill, extract_document_info has no description of expected encoding, max size, or supported formats. LLMs may pass invalid data.
No error handling or recovery guidance visible in tool definitions. Tools lack descriptions of failure modes, retryability, or next steps when calls fail. Agents cannot self-correct.