FeelConnect▍
ProductsDocsPricingChangelog
Sign inGet API keys

Search documentation

Fuzzy search across every docs page and section

Platform/MCP and AI agents

Getting started

  • Overview
  • Quickstart
  • Authentication
  • Environments and sandbox
  • Link
  • Going live

API reference

  • Link tokens
  • Institutions
  • Items
  • Accounts
  • Transactions
  • Identity
  • Income
  • Assets
  • Investments
  • Liabilities
  • Recurring
  • Categories
  • AI products

Platform

  • Webhooks
  • Errors
  • Rate limits and quotas
  • SDKs
  • GraphQL
  • MCP and AI agents
  • Changelog
  1. Docs
  2. Platform
  3. MCP and AI agents
PreviousGraphQLNextChangelog

On this page

  • Endpoint
  • Available tools
  • Authorization model
  • Discovery for AI tools
FeelConnect▍

The developer-first financial data platform.

All systems operational

Products

Universal LinkData APIsAI enrichmentPricing

Developers

DocumentationAPI explorerDashboardGet API keys

Company

HomeChangelogfeels.moneySign in

Legal

PrivacyTermsSecurityData processing
© 2026 FeelConnect. All rights reserved./v1 · FC-Version 2026-07-01

Platform

MCP and AI agents

Connect AI agents to FeelConnect over the Model Context Protocol: the /api/mcp endpoint, available tools, and key-based authorization.

FeelConnect ships a Model Context Protocol (MCP) server so agents — Claude, IDE assistants, and your own — can call the platform as tools instead of hand-writing HTTP. The server exposes the same data and AI products as /v1, with the same key-based authorization and metering.

Endpoint

POST/api/mcp

Transport is Streamable HTTP per the MCP specification. Authorize with a normal secret key; the key's environment decides whether tools touch sandbox or live data — point agents at fc_sk_test_ keys unless you know exactly what you are doing.

claude mcp add feelconnect https://api.feelconnect.dev/api/mcp \
--transport http \
--header "Authorization: Bearer fc_sk_test_..."

Available tools

ToolMaps toDescription
search_institutions/v1/institutions/searchFind institutions by name and type.
create_link_url/v1/link/token/createHosted Link URL for a human to complete in a browser.
exchange_public_token/v1/link/token/exchangepublic_token → permanent access_token.
get_item/v1/items/getItem status and consented products for an access token.
list_accounts/v1/accounts/getAccounts with balances.
get_transactions/v1/transactions/getDate-ranged transactions, up to 250 per call.
sync_transactions/v1/transactions/syncCursor-paged transaction changes.
get_recurring/v1/recurring/getRecurring streams and subscriptions.
get_investments/v1/investments/holdings/getHoldings with securities.
get_liabilities/v1/liabilities/getLoan terms and APRs.
ai_categorize/v1/ai/categorizeCategorize raw transactions.
ai_cashflow_forecast/v1/ai/cashflowDaily cash-flow projection.
ai_health_score/v1/ai/health-scoreExplainable financial health score.
ai_financial_summary/v1/ai/financial-summaryNarrative summary of the last 30 days.
ai_purchase_analysis/v1/ai/purchase-analysisBuy, wait, or skip — with opportunity cost.

Tools that read item data take an access_token argument, exactly like the REST endpoints. The server never mints tokens without a human: create_link_url returns a URL the person opens themselves, so credentials and consent stay in the Link flow.

Authorization model

  • The Authorization: Bearer fc_sk_… header on the MCP connection scopes every tool call to your application and environment.
  • Tool calls are metered as API requests (AI tools also as AI requests) and appear in Dashboard → Logs like any /v1 traffic.
  • Give agents a dedicated key so you can watch and revoke their access independently in the dashboard.
  • Access tokens passed as tool arguments obey the usual binding rules — an agent cannot read items that do not belong to your app + environment.

Agent safety

Treat an agent like any other production client: test keys while developing, least-privilege products in link tokens, and a dedicated revocable key. Never paste live access tokens into prompts you do not control.

Discovery for AI tools

  • /llms.txt — a machine-readable map of these docs for LLM crawlers; /llms-full.txt is the endpoint-by-endpoint expansion.
  • /openapi.json — the complete OpenAPI 3.1 description of /v1, including the webhook payloads.
  • /.well-known/mcp.json — the MCP discovery descriptor: endpoint, transport, auth, and every tool with its arguments.
  • /.well-known/ai-plugin.json — the agent manifest (what this API returns, and how to authenticate).
  • A GET to the MCP endpoint returns the same descriptor, so probing the URL answers rather than 405s.
  • These docs are static HTML — agents can read any page directly.