Connect Claude, ChatGPT, Copilot and other MCP tools to Smart OSH

SmartOSH exposes a remote MCP server that you can connect to Claude, ChatGPT, or another MCP-compatible tool to query your information with AI assistance.

Prior knowledge

This guide assumes that you already understand how your MCP client works (Claude, ChatGPT, Copilot, etc.) and that you are familiar with the concept of MCP. SmartOSH does not provide support for using these tools beyond the help necessary to establish the connection.

What is MCP

MCP (Model Context Protocol) is an open standard that allows AI assistants, such as Claude or ChatGPT, to securely connect to external applications and data. Instead of manually copying and pasting information, the assistant connects to the MCP server and can access the data the application makes available.

SmartOSH has its own MCP server, so you can add it as a custom connector in Claude, ChatGPT (or another compatible tool) and work with your information directly from the assistant.

MCP function is optional

The MCP connection is an optional feature and, for security reasons, it is disabled by default (explicit opt-in). If the feature is not enabled, the connector will not be able to connect.

Activation is not done from the settings: it is the consultant or support team who enables it at the client’s request. For security, the SmartOSH team verifies the request before activating the service.

Once the service is activated, the specific tools —such as sending emails— and the authorised domains for that sending are configured in the settings, in the same way as the rest of SmartOSH’s AI functionalities.

What SmartOSH’s MCP offers

SmartOSH’s MCP server goes beyond exposing data: it incorporates several tools, predefined prompts, and specific PRL knowledge so the AI assistant can work more usefully and guided. Among other capabilities, it offers:

• Information retrieval: querying accidents, reports, and documents, among other data.
• Sending emails directly from SmartOSH (subject to the configured authorised domains).
• Predefined prompts for common tasks, which facilitate obtaining consistent results without having to write instructions from scratch.
• Specific accident investigation methodology and other specialised PRL knowledge incorporated into the service itself.

Note

The set of available tools may expand over time and depends on the features enabled for your organisation.

SmartOSH MCP server URLs

Use the URL corresponding to the environment you want to work with:

• Production:

https://backend.my.smartosh.com/mcp

• Preproduction:

https://backend.pre-my.smartosh.com/mcp

Caution

These are the generic URLs. Normally, each client has their own custom URL in the format:

https://MYCOMPANY.backend.my.smartosh.com/mcp

Where MYCOMPANY is your organisation’s custom subdomain. If you do not know this subdomain, you should request it from your consultant or support before configuring the connector.

How to connect Claude (custom connector)

To add a custom connector in Claude you need a remote MCP server to connect to; in this case, one of the SmartOSH URLs indicated above. The steps vary depending on your Claude plan.

Individual plans (Free, Pro or Max)

1. Go to Customize > Connectors in settings, or click the «+» button next to Connectors.
2. Click Add custom connector to create a new one.
3. Enter a name to identify it (for example, SmartOSH).
4. Add the SmartOSH remote MCP server URL (the one your consultant or support has provided).
5. (Optional) Click Advanced settings to specify an OAuth Client ID and an OAuth Client Secret for your server.
6. Click Add and follow the connection process (authentication), just like with directory connectors.

Team or Enterprise plans

In these plans, before members can configure custom connectors, an Owner or Primary Owner must first add the connector to the organisation:

1. The Owner or Primary Owner goes to Organization settings > Connectors, clicks Add and adds the SmartOSH remote MCP server URL.
2. Each user then finds the connector in the list (it will appear with a «Custom» label) and clicks Connect to authenticate and start using it.

Availability by plan

Custom connectors via remote MCP are available in Claude, Cowork and Claude Desktop for the Free, Pro, Max, Team and Enterprise plans. Free users are limited to a single custom connector, while Pro, Max, Team and Enterprise support multiple.

How to connect ChatGPT (custom connector)

In ChatGPT, custom MCP connectors are managed through Developer mode.

Enable Developer mode

1. Open ChatGPT on the web and sign in with a compatible plan account.
2. Go to Settings > Connectors —depending on your version, this may appear as Apps.
3. Enter Advanced settings; you may need to scroll down to find it.
4. Toggle Developer mode on.

Create the connector

1. In Settings > Connectors (or Apps), click Create next to Advanced settings. The button only appears if Developer mode is enabled.
2. Enter a name to identify it (for example, SmartOSH).
3. (Optional) Add a description, for example “Access to SmartOSH data”.
4. In MCP Server URL, paste the SmartOSH remote MCP server URL.
5. In Authentication select the appropriate method (usually OAuth) and complete the authentication process.
6. Save the connector. It will appear in your connector list (in Drafts while being configured).

Use it in a conversation

1. Open a new conversation and click the «+» icon in the composer.
2. Select Developer mode and activate the SmartOSH connector as a source.

Availability by plan in ChatGPT

Full MCP connector support in ChatGPT is available on the Plus, Pro, Business, Enterprise and Education plans (not on the Free plan) from ChatGPT web. On Plus and Pro tools are usually read-only, while on Business and Enterprise/Education write actions are supported, typically with administrator authorisation.

How to connect Microsoft Copilot

In the Microsoft ecosystem, MCP servers are connected through Copilot Studio, where they are added as a tool to an agent.

1. Enter Copilot Studio and open (or create) the agent to which you want to add the connection.
2. Go to the Tools tab and click Add a tool > New tool > Model Context Protocol.
3. Fill in the wizard fields:
   • Server name: for example, SmartOSH.
   • Server description: a brief explanation of its purpose, as the orchestrator uses this to decide when to call the server.
   • Server URL: the SmartOSH remote MCP server URL.
4. Configure the appropriate authentication (usually OAuth 2.0) and complete the wizard. Copilot Studio automatically detects the available tools on the server.

Note

Access to MCP servers in Copilot Studio is governed by your organisation’s Power Platform data policies.

How to connect Gemini

In Gemini the usual way to connect remote MCP servers is via the Gemini CLI (aimed at technical profiles). With the CLI installed, add the server with:

gemini mcp add --transport http SmartOSH <MCP-server-URL>

If the server requires header authentication, you can add it with -H:

gemini mcp add --transport http --header "Authorization: Bearer <token>" SmartOSH <MCP-server-URL>

This saves the configuration in ~/.gemini/settings.json (user scope) or .gemini/settings.json (project scope). Within Gemini CLI you can run /mcp to see the configured server and its tools.

Note

In Google Cloud corporate environments, MCP connectors can also be managed from the Gemini Enterprise Agent Platform, specifying the server name, URL, HTTP transport and authentication data (Google Cloud credentials or OAuth Client ID/Secret).

Other MCP-compatible clients

SmartOSH’s MCP server follows the MCP standard, so it can connect from any compatible client (for example, editors and development tools like Cursor, VS Code or Cline, among others). Although the interface changes depending on the tool, generally you only need these details:

• The SmartOSH remote MCP server URL (production or preproduction).
• The authentication method (for example, OAuth, with Client ID and Client Secret if applicable).

Tip

If your tool does not appear on this page, consult its documentation on how to add a remote MCP server and use the URL and authentication data provided by your consultant or support.

Responsibility and data protection

Through the MCP service, SmartOSH limits itself to exposing data and tools. The processing of information by the AI is performed by the external engine or tool you choose to connect (Claude, ChatGPT, Copilot, Gemini or another), outside SmartOSH’s control.

During MCP use, SmartOSH does not use its own AI engines. Therefore, token consumption (and any associated cost) is borne by the client and the AI tool being used.

Therefore, the following aspects are the responsibility of the client and the AI tool used:

• The legal aspects of AI use.
• The geographical area where the AI runs and data is processed.
• The data protection implications and compliance with applicable AI regulations.

Personal data in text fields

SmartOSH makes its best effort to pseudonymise data related to employee names. However, descriptions and free text entered by users in description fields may contain personal data.

The regulatory control over where and how this data is processed remains the responsibility of the client who activates the MCP service. Before enabling it, ensure that the chosen AI tool and its processing region comply with your legal and data protection obligations.