# How to Connect Tessitura to Claude, ChatGPT and Gemini with MCP

> MCP lets Claude, ChatGPT and Gemini work with live Tessitura data. What it is, what staff can ask, and how to roll it out safely, starting read-only.

- URL: https://digitalartsmediaservices.com/blog/connect-tessitura-to-claude-chatgpt-gemini-mcp
- Published: 2026-09-22
- Author: Digital Arts Media Services (Washington, DC)
- Topic: Tessitura

## Key takeaways

- The Model Context Protocol (MCP) is an open standard that lets AI assistants such as Claude, ChatGPT and Gemini use outside tools and data through a server you control.
- An MCP server for Tessitura sits between the AI and the Tessitura REST API, so the AI only gets the specific actions and fields you choose to expose.
- The safest rollout starts read-only, with a handful of well-defined tools, a dedicated Tessitura service user, and testing in your Tessitura test environment.
- Good servers control access per tool, per role and per field, redact sensitive data, require approval for any change, and log every call.
- One governed MCP server can serve staff in their AI assistants, your internal apps and a patron-facing chatbot, so there is one integration to maintain.

To connect Tessitura® to Claude, ChatGPT or Gemini, you put an **MCP server** in between. The server exposes a short list of approved Tessitura actions, such as "check availability" or "summarize this week's sales", through the Model Context Protocol. Your staff add it to their AI assistant, sign in, and ask questions in plain English, answered from your live data.

That is the whole idea. The rest of this guide covers what MCP is, what your teams can do with it, how a safe server is put together, and how to roll it out without putting patron data at risk.

## What is the Model Context Protocol (MCP)?

The Model Context Protocol is an open standard for connecting AI assistants to outside tools and data. Anthropic introduced it in late 2024, and it has since been adopted across the industry, so the same connection can work with many different AI tools.

A useful way to picture it:

- **The AI assistant** (Claude, ChatGPT, Gemini, or a developer tool) is the one your staff talk to.
- **The MCP server** is a small service you control. It publishes a list of *tools*, each with a name, a description and the inputs it accepts.
- **Your system** (here, Tessitura) is only ever reached by the server, never by the AI directly.

When someone asks a question, the AI decides which tool to call, the server checks whether that person is allowed to use it, runs the request against Tessitura, and sends back only the fields it is permitted to share. The AI then writes the answer.

![Diagram: AI assistants such as Claude, ChatGPT, Gemini and developer tools send tool calls to your MCP server. The server allows approved tools only, checks sign-in and roles, redacts sensitive fields, requires approval before any write and logs every call, then makes API requests to the Tessitura REST API. The AI never calls Tessitura directly.](https://digitalartsmediaservices.com/media/blog/connect-tessitura-to-claude-chatgpt-gemini-mcp/how-mcp-works.webp "The MCP server sits between your staff's AI tools and Tessitura, and decides what each person can see and do.")

Because MCP is a standard, you build the integration once. When your team changes AI tools, or a new one comes along, the server keeps working.

## Why connect Tessitura to an AI assistant?

Your staff are probably already using AI assistants for drafting and research. What those assistants can't do is see your season, your sales or your patrons. So people export reports, paste rows into a chat window, or simply go without. Pasting exports is slow, and it also scatters patron data into places nobody is tracking.

An MCP server fixes both problems. The AI gets live, current answers from Tessitura, and your organization decides exactly what it can reach.

## What can each team ask?

Here are the kinds of questions that become a single sentence once Tessitura is connected:

**Box office**
- "What's still open in the orchestra for Saturday's matinee?"
- "Find this patron's upcoming orders and resend their confirmation."

**Marketing**
- "Which productions are pacing behind last season at two weeks out?"
- "How many first-time buyers from the fall haven't returned yet?"

**Development**
- "Brief me on this donor before tomorrow's lunch."
- "List lapsed donors at the $1,000 level and above, and when we last spoke to each."

**Leadership**
- "Summarize this week's ticket revenue against budget by production."
- "Compare subscription renewals with the same point last year."

**Developers**
- In a coding assistant: "Check how price types are set up for this production, then update our template to match."

None of these require a new report to be built. They do require the right tools on the server, which is where the design work goes.

## How is an MCP server for Tessitura built?

A trustworthy server has five building blocks.

### 1. Tools on the Tessitura REST API

Each tool is a narrow, well-named action built on the Tessitura REST API: `get_season_calendar`, `get_performance_availability`, `get_sales_summary`, `search_constituents`, `create_constituent_note`. Narrow tools are easier for the AI to use correctly and much easier for you to govern than one tool that can "query anything".

### 2. Authentication that fits the audience

You can run the same server, or separate servers, for different audiences:

| Audience | How they connect | Typical access |
| --- | --- | --- |
| The public | No sign-in | Read-only season, performance details, availability and pricing |
| Your staff | Sign-in through your identity provider (SSO / OAuth) | The tools and data their role allows |
| Your systems | API keys | Internal apps, automations, AI agents and a patron chatbot |

### 3. Read, write, or both

Read tools answer questions. Write tools change things, such as adding a note, updating contact details or holding seats. Write access should be opt-in, limited to specific roles, and able to require a person's approval before anything is saved in Tessitura.

### 4. Controls per tool, per role and per field

This is where an MCP server earns its keep. A box office role might see sales and patron orders but not giving history. A marketing role might search constituents with personal details redacted. Development might see contributions but not be able to hold seats. Sensitive fields, like payment details, birthdays or private notes, can be stripped out by the server so they never reach an AI model at all.

### 5. Logging and limits

Every tool call should be logged with who asked, what ran and what changed. Rate limits and query caps protect your Tessitura instance from runaway requests, whether those come from a confused AI or a curious user.

## How should we roll it out?

A step-by-step rollout keeps risk low and builds trust with staff, IT and leadership.

![Seven-step rollout. Plan: discovery, design tools and permissions, create a dedicated service user. Build: test in your Tessitura test environment. Launch: a read-only pilot, then train and widen access. Extend: add write actions last, by role, with a person's approval.](https://digitalartsmediaservices.com/media/blog/connect-tessitura-to-claude-chatgpt-gemini-mcp/rollout.webp "Start read-only with a pilot group, and add write actions last.")

1. **Discovery.** List the real questions each team asks today and how they get answers now. These become your first tools.
2. **Design tools and permissions.** Decide which tools exist, which roles can use each one, and which fields stay private. Write this down; it becomes your policy.
3. **Create a dedicated Tessitura service user.** Give the server its own user with only the permissions it needs, so it is easy to audit and easy to switch off.
4. **Build and test in your Tessitura test environment.** Nothing touches live data until the tools behave as expected.
5. **Launch read-only with a pilot group.** Pick a few people from different teams, connect their AI tools, and gather feedback for a few weeks.
6. **Train, then widen access.** Show staff what to ask, what the AI can't see, and how to check an answer. Then add more people and more tools.
7. **Add write actions last.** Only where they clearly save time, limited by role, with approval required.

If your organization is also writing an AI policy, do it alongside this rollout. Our guide to [writing an AI policy for an arts nonprofit](https://digitalartsmediaservices.com/blog/ai-policy-for-arts-nonprofits) covers what to include.

## What are the security considerations?

Connecting AI to a system full of patron and donor records deserves care. The main questions to answer before launch:

- **Which AI providers are approved?** Data returned by a tool goes to the AI provider your staff are using, under that provider's terms. Business and enterprise plans from the major providers generally exclude your data from model training. Check the current terms for each product your team uses.
- **What leaves the server?** Return only the fields a task needs. Redact by default and allow fields deliberately.
- **Who can connect?** Staff should sign in with their work accounts so access follows their role and ends when they leave.
- **What can change?** Keep writes off until you need them, and require approval when you turn them on.
- **What gets logged, and who reviews it?** Logs are only useful if someone looks at them.
- **Where does the server run?** In your own cloud or hosted by a partner, with monitoring and updates as Tessitura and the AI tools change.

We go deeper on each of these in our [privacy and security checklist for using AI with Tessitura data](https://digitalartsmediaservices.com/blog/tessitura-ai-data-privacy-security-checklist).

## Should we build our own MCP server or work with a partner?

If you have developers who know the Tessitura REST API, building a small read-only server is a reasonable project, and the protocol itself is well documented. The harder parts are rarely the first tool. They are the permission model, field redaction, sign-in through your identity provider, logging, testing against real Tessitura data shapes, and keeping it all working as the AI tools evolve.

A few questions help decide:

- Do we have in-house time to maintain this for years, not weeks?
- Do we know the Tessitura API well enough to write efficient, safe queries?
- Can we support SSO, audit logs and approvals from day one?
- Who will answer when staff ask, "Why did the AI say that?"

If your team wants to build it themselves, a developer workbench like our [DevHub for Tessitura](https://digitalartsmediaservices.com/services/tessitura-devhub) helps you explore the API, test calls on a demo venue and publish your work as MCP tools. If you'd rather have it built and run for you, that is what our [MCP Servers for Tessitura](https://digitalartsmediaservices.com/services/tessitura-mcp-server) service does.

## How does this fit with a patron chatbot and AI agents?

The same governed tools can power more than staff questions. A [box office chatbot](https://digitalartsmediaservices.com/blog/tessitura-ai-chatbot-box-office-guide) can use the public, read-only tools to answer patrons on your website. [AI agents](https://digitalartsmediaservices.com/services/ai-agents) can use staff tools to prepare donor briefings or weekly sales summaries for someone to review. One server, one set of rules, one integration to maintain. Our [AI for Tessitura](https://digitalartsmediaservices.com/ai-for-tessitura) page shows how the pieces fit together.

## Ready to connect Tessitura to your team's AI tools?

If you'd like to see what your staff could ask on day one, we're happy to help. We'll map your first read-only tools, the roles that should use them and the fields that should stay private. [Book a free consultation](https://digitalartsmediaservices.com/contact), and we'll reply within 24 hours.

## Frequently asked questions

### What is an MCP server for Tessitura?

It is a small, secure service that exposes a defined set of Tessitura actions, such as checking availability or summarizing sales, to AI assistants through the Model Context Protocol. The AI never calls the Tessitura API directly. The server decides what each user can see and do, and it logs every request.

### Can Claude, ChatGPT and Gemini all use the same MCP server?

Yes. MCP is an open standard, so one server can work with any AI tool that supports connecting to MCP servers. Support and setup differ by product and plan, so check each tool's current connector settings before you roll it out to staff.

### Does connecting AI to Tessitura put patron data at risk?

It depends on how the server is designed. A well-built server only returns the fields a task needs, redacts sensitive ones, limits tools by role and logs every call. Data goes only to the AI provider your organization chooses, under that provider's terms, so use business or enterprise plans that exclude your data from model training.

### Should an AI assistant be allowed to change data in Tessitura?

Start without write access. Once your team trusts the read-only tools, you can add narrow write actions, such as adding a constituent note, limited to specific roles and requiring a person to approve each change before it is saved.

### What do we need on the Tessitura side to get started?

Access to the Tessitura REST API, the same API your website and TNEW integrations use, plus a dedicated service user with appropriate permissions. Building and testing against your Tessitura test environment first keeps live data safe while you learn.

---

Tessitura® and TNEW® are registered trademarks of Tessitura Network, Inc. Digital Arts Media Services is independent and not affiliated with, endorsed by, or sponsored by Tessitura Network, Inc. ChatGPT is a trademark of OpenAI; Anthropic and Claude are trademarks of Anthropic, PBC; Gemini is a trademark of Google LLC. Product names are used for identification only.