MCP Server (AI Access)

Important

Beta Feature — MCP Server (AI Access) is currently in beta and is being rolled out to databases over the coming weeks. It is not yet available everywhere: if you do not see the AI Features section in the Admin menu, the feature has not reached your database yet. Tools, settings, and screens described in this section may still change during the beta.

The TouchPoint MCP server lets AI chat agents — such as Claude, ChatGPT, and Cursor — work directly with your church’s TouchPoint data. MCP (Model Context Protocol) is an open standard for connecting an AI assistant to an outside system through a defined set of tools. Instead of pasting exports into a chat window, your staff can ask questions in plain language and the agent calls TouchPoint on their behalf: “Who joined our Young Adults involvement in the last month?”, “Log a call touchpoint on this person”, “Build a Search Builder query for members with no attendance since June.”

Each church enables the MCP server independently, and nothing is exposed until an administrator turns it on. Every connection is tied to a specific TouchPoint user, so an agent can only see and do what that user could already see and do in TouchPoint — with extra restrictions on top.

Your church’s endpoint is:

https://mychurch.tpsdb.com/api/mcp

Replace mychurch with your church’s TouchPoint subdomain.

What It Takes to Get Started

There are three layers to set up, in this order:

  1. TouchPoint Support enables the AI Features for your database. The MCP feature is hidden until the SupportMcpFeatures setting is turned on for your database. Contact TouchPoint Support to request it.

  2. An administrator enables and configures the MCP server on the MCP Server (AI Access) page, and grants the McpAccess role to the staff who should be able to connect. See Configuring the MCP Server.

  3. Each staff member connects their own AI client using a Personal Access Token issued to their own TouchPoint user account. See Connecting an AI Client.

Important

When an AI agent calls an MCP tool, the AI provider (Anthropic, OpenAI, etc.) receives the data returned by that tool and handles it under that provider’s terms — which may include training public models unless the client is configured to opt out. Review the data-handling settings of every client you allow your staff to connect before you turn the MCP server on.

How Access Is Controlled

Several independent gates apply to every MCP tool call. All of them must pass:

  • MCP Server: Enabled — the per-church switch on the admin page. When off, every connection is rejected.

  • A valid Personal Access Token — the agent authenticates as the TouchPoint user who owns the token.

  • The McpAccess role — the token’s user must hold it, or no tool will run.

  • The user’s normal TouchPoint roles — the same role checks that apply in TouchPoint apply here. Giving tools require a Finance role, roster tools require Attendance, ManageOrgMembers, OrgLeadersOnly, or Admin, and so on.

  • Write actions — creating or updating data requires the Allow Write Actions setting in addition to everything above. It is off by default.

  • Field-level masking — contact and demographic fields come back as [redacted] unless the user also holds McpViewContact or McpViewDemographics.

Every tool call is written to the activity log with the user who made it, the tool that ran, and a summary of what was asked, so you always have an audit trail of what the AI agents did.

See also

Personal Access Tokens