Skip to main content

Core web: How do I add the Core MCP as a custom connector?

How to add Core as a custom connector in Claude or ChatGPT using an OAuth connection.

C
Written by Chris

Core exposes a Model Context Protocol (MCP) server that lets an AI assistant connect to a user's Core wallet, read balances, and, with the user's approval, sign and broadcast transactions, subject to the MCP configuration a user sets in Core web's MCP policy controls.

To configure the MCP server, see this article. To view the MCP server's available toolset, see this article.

WARNING: Connecting Core as a custom connector allows an AI assistant to request transaction signatures on the user's behalf. Only connect trusted clients, review every transaction before approving it, and confirm the account's spend policy is configured before use.

NOTE: Core uses an OAuth connection. When the connector is added, a sign-in window opens so the user can authorize access to their Core wallet. The user's private keys and recovery phrase are never shared with the AI client.

Before starting

A user needs the following before adding the connector:

  • A Core seedless wallet with at least one account.

  • The Core MCP server URL: https://mcp.core.app/mcp

  • An AI client that supports custom remote MCP connectors.

NOTE: In this article, Claude and ChatGPT are shown as examples of AI clients that support custom connectors. However, not all plans may be eligible. Please refer to Anthropic and OpenAI's documentation to see if a plan supports custom connectors.

Add Core as a connector in Claude

  1. In Claude, open Settings.

  2. Select Connectors.

  3. Select Add custom connector.

  4. Enter a name for the connector, such as Core.

  5. In the URL field, enter the Core MCP server URL: https://mcp.core.app/mcp

  6. (Optional) Under Advanced settings, enter an OAuth Client ID and OAuth Client Secret only if these were provided. Core supports OAuth sign-in, so most users can leave these blank.

  7. Select Add.

  8. On the Core connector, select Connect.

  9. In the OAuth window, sign in to Core and authorize access.

Once access is authorized, the Core tools become available to Claude.

NOTE: Core only signs transactions that pass the user's configured MCP policy. Depending on the AI client and workspace settings, the client may also ask the user to approve tool calls before they run. Users should configure accounts, networks, spend limits, and allowed recipients before enabling transaction-capable use.

Add Core as a connector in ChatGPT (web only)

NOTE: Custom MCP connectors in ChatGPT require Developer mode, which is available on eligible plans. If the Developer mode toggle is not visible, a workspace admin may need to enable it in the workspace settings.

First, enable Developer mode:

  1. Open Settings and select Security & Login.

  2. Toggle Developer mode on.

Then add the connector:

  1. In Settings, select Plugins.

  2. Select + in the top right to open the New App menu.

  3. Enter a name, such as Core, and a short description.

  4. In the MCP server URL field, enter: https://mcp.core.app/mcp

  5. Set Authentication to OAuth.

  6. Select Create.

  7. Complete the OAuth sign-in to authorize access to Core.

To use the connector in a conversation, enable Core as a source, then confirm each tool call Core proposes before it runs.

NOTE: Core only signs transactions that pass the user's configured MCP policy. Depending on the AI client and workspace settings, the client may also ask the user to approve tool calls before they run. Users should configure accounts, networks, spend limits, and allowed recipients before enabling transaction-capable use.

Disclaimer

The Core Knowledge Base, including all the Help articles on this site, is provided for technical support purposes only, without representation, warranty, or guarantee of any kind, and is not technical, investment, financial, accounting, tax, or legal advice. The Core Wallet MCP Server enables an authorized AI agent to sign and broadcast transactions from your wallet. The spend-policy controls described above are provided on a best-efforts basis, apply only to the transaction types and networks specified, and do not guarantee against loss — including loss resulting from agent behavior, network fees, failed or unavailable simulation, or transactions falling outside the scope of a given control. You are responsible for the accounts, recipients, networks, limits, and agents you choose to enable. Examples are illustrative only. Please review this Notice and the Terms of Use.


For any additional questions, please view our other knowledge base articles or contact a Product Support team member via the chat button. Examples are for illustrative purposes only.

Did this answer your question?