What is the Candor Finance MCP server?
Candor Finance is Candor's official remote MCP server for personal finance, registered as money.candor/candor-finance on the official MCP Registry and served at https://api.candor.money/mcp over Streamable HTTP with OAuth. Any agent that supports remote MCP can connect without installing anything locally.
The official MCP Registry record points to the same endpoint and to the public package on GitHub, so an agent platform or reviewer can inspect the declared server, the finance skill, and the connection metadata before installing.
The server pairs with the candor-finance skill. MCP carries the tools and records; the skill teaches the agent the methods, such as checking coverage before a total, investigating a possible duplicate charge, or following up on a missing refund. Supported installs keep both on matching versions.
What does an agent get from Candor's MCP server?
A workspace of its own. Your agent keeps goals and budgets there with their history, plus notes and decisions with their reasons, so the next conversation starts from them. Every week Candor's review raises up to five areas worth talking through and up to three gaps worth filling, each pointing at its records. Your agent reads them with candor_open, then pulls the exact records and figures.
The review covers what a careful advisor would raise: a price increase, a duplicate charge, cash sitting idle while a card charges interest, budget drift, missing or changed income, a debt deadline, goal pace, and questions with no single transaction behind them, such as an employer match left on the table or a tax deadline. If nothing qualifies, the list is empty.
Totals, budget pace, and variance are calculated from your records, so the agent reports a figure instead of predicting one. candor_query and candor_get return bounded records with freshness and coverage, candor_visualize draws a dashboard panel where the agent supports it, and candor_write saves approved state. Candor is not a licensed advisor: the agent makes suggestions and you decide.
Which AI agents work with Candor's MCP server?
Claude, ChatGPT, Grok Bot, Meta Muse, Claude Code, Codex, Gemini CLI, OpenClaw, Hermes, and any agent that can load the finance skill and use remote MCP with OAuth. Agents that cannot use MCP can use the Candor CLI instead. Every agent opens the same workspace, so plans made in one are there in the next.
Your agent reads the review when you ask, or on a schedule if it supports scheduled tasks. Scheduling belongs to the agent, not to Candor, so set it up there and have the agent confirm it.
How does an agent sign in to Candor?
Through OAuth 2.1 with dynamic client registration. Your MCP client handles the authorization challenge, and you sign in and approve that client on Candor's own page. Bank credentials never pass through the agent or the chat, and you can revoke a client at any time.
Accounts connect separately, through bank-grade secure sign-in in Candor's web app. Bank logins, access tokens, full account numbers, and verification codes never belong in a conversation.
Hosted OAuth clients work too. When a client omits the resource parameter, Candor binds the MCP resource before consent; tokens stay limited to their authorized audience, and hosted browser callbacks do not need access to the agent's computer.
Can an agent move money through Candor?
No. Candor's MCP server has no tool to move money, pay a bill, trade, or cancel a service. The agent can save workspace state you approve, such as a budget, goal, or note. Anything outside Candor, the agent does through its own tools with your approval. Every read and change is recorded with its reason.
Candor never sells your data or trains AI models on your financial information. You can export what Candor holds, disconnect an account, revoke an agent, or delete your data from account settings.
How does Candor's server compare with other finance MCP servers?
Most finance MCP servers let your agent read, and sometimes edit, the records in an app or a bank feed. Candor's gives your agent a plan of its own: goals and a budget it sets up from your accounts and keeps current, figures Candor calculates from your records, decisions kept with their reasons, and a weekly review of what needs attention.
Common questions
- What is a personal finance MCP server?
- A Model Context Protocol server that gives an AI agent tools for financial data, such as accounts, transactions, and budgets. The agent calls those tools instead of you pasting statements into a chat. Candor Finance is a hosted one: agents connect over remote MCP with OAuth.
- Does Claude or ChatGPT need anything else to use Candor?
- In ChatGPT, install Candor Finance from the plugin directory; it brings the finance skill with it. In Claude, add Candor Finance from the connector directory, then paste Candor's one-line setup message so the agent adds the finance skill. Then approve its access on Candor's page. The setup pages cover each agent step by step.
- How much does the Candor Finance MCP server cost?
- The server is included in a Candor subscription. Candor starts at $8.25 a month on the annual plan ($99 a year), or $14.99 month to month, and the first 7 days are free.
- Do I need a Plaid account or my own server?
- No. Candor runs the server and holds the Plaid connection, so there are no API keys or access tokens to manage. You link each bank in Candor and your agent signs in with OAuth. A self-hosted Plaid MCP server needs your own Plaid account and a process you keep running.
- Where is the server officially listed?
- The canonical record is money.candor/candor-finance on the official MCP Registry. It points to the production endpoint api.candor.money/mcp and the public package on GitHub.