AI Assistant Compatibility · reviewed 19 July 2026
Compatibility is validated for the AI assistant, plan, and environment you intend to use.
Systems Place can connect Managed AI Analytics to compatible off-the-shelf AI assistants. This reference focuses on ChatGPT and Claude. Availability and behaviour vary by product surface, plan, workspace settings, authentication, network path, and the tools required.
The short answer
Provider support is a starting point, not a guarantee for every workspace.
Model Context Protocol, or MCP, allows an AI assistant to call tools supplied by another service. Support can still vary by product surface, plan, workspace policy, authentication method, network access, and whether the tools only read data or also propose changes.
Systems Place reviews those dependencies during the assessment and confirms the selected configuration in the environment intended for production. The dated matrix below focuses on managed business deployments.
Dated compatibility matrix
Start from provider support, then validate the exact configuration.
This matrix summarises the documented requirements for the assistant products most relevant to a managed business deployment.
| Assistant and plan | Supported surface and administration | Connection and authentication | Network constraint | Material limitations and status |
|---|---|---|---|---|
| ChatGPT Business | ChatGPT web. An admin or owner enables developer mode, creates the custom app and publishes it to the workspace. | Remote MCP app. The app setup selects the available authentication method; OAuth refresh behaviour must be tested when used. | ChatGPT must be able to reach the remote MCP server. Any private-network or on-premises connection method must be verified for the customer environment. | The developer-mode deployment path is documented for ChatGPT web. Full MCP tools, including write and modify actions, are rolling out in beta; each enabled action and customer use case needs review and acceptance. |
| ChatGPT Enterprise / Edu | ChatGPT web. Admins and owners publish apps; role-based access and action controls are available for the workspace. | Remote MCP app with the authentication supported by the selected configuration. User access and each enabled action remain workspace decisions. | ChatGPT must be able to reach the remote MCP server. Any private-network or on-premises connection method must be verified for the customer environment. | The developer-mode deployment path is documented for ChatGPT web. Enterprise and Edu add role-based access controls. Full MCP tools, including write and modify actions, are rolling out in beta and require customer-specific acceptance. |
| Claude Team / Enterprise | Remote custom connectors are documented for Claude, Cowork, and Claude Desktop. An organisation owner adds the connector and each user connects it individually. | Remote MCP connector, typically using OAuth where authentication is required. Owners may supply an OAuth client ID and secret when the server design needs them. | Remote connector traffic originates from Anthropic cloud infrastructure. The service must be reachable from the documented IP ranges; a user VPN or desktop network path is not sufficient. | Custom connectors are in beta and are not verified by Anthropic. Tool scope, approval behaviour, and the selected Claude product must pass customer-specific acceptance. |
Provider sources reviewed 19 July 2026: OpenAI: Apps in ChatGPT; OpenAI: developer mode and MCP apps in ChatGPT; Claude: custom connectors using remote MCP. OpenAI also documents read and fetch MCP connections for Pro users, while full MCP remains limited to Business, Enterprise, and Edu. Anthropic documents remote custom connectors for Free, Pro, and Max, with Free limited to one. Individual plans are useful for evaluation but do not replace managed-workspace ownership and policy decisions. Provider behaviour and availability may change after this date.
Validation path
Compatibility moves from documented assumption to tested acceptance.
The depth of testing is proportionate to the question, source sensitivity, AI assistant, and service responsibilities.
Identify
Design
Test
Record
Recheck
Compatibility must be maintained. A passing result applies only to the tested combination and date, not to every plan or future provider release.
What validation means
Validate the configuration buyers will actually use.
Move from general protocol support to a tested plan, product surface, identity path, network path, and tool experience.
Protocol support or a successful developer test.
- The provider mentions MCP in general documentation
- A different plan or product surface can connect
- A developer can invoke one tool in a separate environment
- The connection works without the customer’s identity and policy controls
The selected configuration passes agreed acceptance tests.
- The named plan, surface, workspace settings, and users are recorded
- Authentication and network behaviour work in the intended environment
- Representative tools, errors, results, and limits are tested
- Owners, date, dependencies, and revalidation triggers are documented
Provider boundary
The assistant provider remains part of the data and control path.
Questions, tool requests, and returned results may be processed by the selected assistant provider according to the customer’s plan, settings, terms, retention choices, and regional availability. The assessment documents this dependency for the selected option.
Terms and settings
Customer provider agreements and workspace controls are reviewed as deployment inputs.
Returned data
Only the data needed for an approved result should be returned, subject to the implemented tool and source limits.
Identity boundary
The available assistant and source identity mechanisms are documented; permission inheritance is never assumed.
Change dependency
Provider releases, settings, terms, and feature availability may trigger review or revalidation.
Common questions
Compatibility questions that need a specific environment to answer.
How is compatibility confirmed across AI assistants?
The target combination is designed and tested explicitly because protocol support, authentication, tool behaviour, product surfaces, plans, settings, and user experience vary.
Does ChatGPT support the required connection for our workspace?
That depends on the exact ChatGPT plan, product surface, workspace controls, administrator configuration, region, authentication, network path, and tools required. These are checked during the assessment and acceptance testing.
Does Claude support the required connection for our organisation?
That depends on the exact Claude plan, product surface, organisation controls, administrator configuration, region, authentication, network path, and tools required. These are checked during the assessment and acceptance testing.
Can the same service target both ChatGPT and Claude?
Potentially, when both selected AI assistants can support the required connection, authentication, tools, result handling, and controls. Each assistant is tested separately, and product-specific adaptation may be required.
Will our existing source-system permissions carry over automatically?
They must not be assumed to. The source, MCP service, and assistant may provide different identity and authorisation capabilities. The implemented enforcement and any gaps are documented for the selected architecture.
When is compatibility rechecked?
It is rechecked before production acceptance and when a material assistant product, plan, workspace setting, authentication method, network path, tool behaviour, source integration, or provider requirement changes.
Validate the intended path
Bring the assistant, plan, source system, and first business question.
The assessment will identify the compatibility assumptions that matter and define how they must be tested before production acceptance.