Orcaset MCP Quick Start
The MCP connector lets you add Orcaset modeling capabilities to Claude, ChatGPT, and many other agents. It provides a secure model execution environment and interactive UI visualizations that you can use with your existing AI providers.
The connector is in early access and subject to frequent improvements and changes.
2. Build
After installing the Orcaset MCP connector, connecting it to your agent, and signing in, you can add workspaces, build and run models, view output, and manage security features.
2.1 Add a workspace
Organize models and analyses with workspaces. Each workspace is a standalone project with its own version control and security settings.
The connector has tools to list, add, and remove workspaces. The agent can list and add new workspaces directly, but deleting a workspace requires manual confirmation.
Create and delete a workspace. This will open a dialog to confirm deletion.
Workspace deletion must be manually approved to prevent accidental removal.
2.2 Build and view models
The connector contains skills that guide the agent through Orcaset modeling best practices. Asking the agent to build or modify an Orcaset model prompts it to review those skills.
Create a simple model for an oil company using the prompt below.
The connector can also show output in formatted browser views. Asking the agent to open a view returns a link that opens in your browser. Access to the linked content requires authentication. You may need to sign in again if the browser session you used to authenticate the connector has expired.
Formatted views present output in familiar statement structures. Specific values and structures will vary because the demo instructions didn’t include concrete details.
Clicking a value highlights its immediate precedents and dependents. Double-clicking a value or pressing the space bar while it is selected opens a dialog with detailed information about the value, formula, and dependency graph.
2.3 Connect data
By default, the model sandbox blocks access to all outside websites (other than the built-in site where files are saved). This safeguard helps prevent accidental or malicious data exfiltration. You can permit access to specific additional sites by modifying the allowlist.
For example, we could update the model to fetch the most recent price for Brent crude from the U.S. Energy Information Administration (EIA). First, add api.eia.gov as a permitted destination in the sandbox. This command will open a dialog.
Agents cannot add new sites to the allowlist autonomously. Each addition must be manually approved.
Like many other APIs, the EIA API is protected by an API key. API keys are long alphanumeric tokens that act like passwords. Since anyone with a token can use it to access protected resources, keys must be stored securely.
Orcaset helps keep keys secure by holding them outside the sandbox and injecting them into web requests at the edge. This prevents keys from leaking to AI agents.
You can get an API key by creating an account on api.data.gov, or you can use the key DEMO_KEY for testing purposes. Ask the agent to add a new API key for api.eia.gov to open the token dialog. Never expose API keys by pasting them into the agent chat. Always use the dialog to securely add keys directly to the Orcaset connector.
Adding an API key automatically adds the associated domain to the allowlist. We manually added it to the allowlist here just for demo purposes.
Type in DEMO_KEY as the value and save.
With the API key added, update the model to determine revenue based on barrels multiplied by the most recent Brent crude price.
Refresh the statement view. Whenever the model runs, it will fetch the most recent Brent crude price from the EIA.
The model now securely fetches live data directly from api.eia.gov. We can use the same pattern to incorporate data from other sources, such as vendor APIs or internal datastores.
3. Deploy
The MCP connector lets firms quickly and securely leverage Orcaset’s data integration and agent-first design. It works with Claude, ChatGPT, and other agent frameworks out of the box without custom development resources.
Connect to existing AI infrastructure using standard MCP integration
View statement dashboards and trace calculations
Keep data secure with networking and secrets management
The demo MCP connector imposes limits on active resources, such as the number and size of workspaces. Reach out to the Orcaset team to increase connector resources or discuss other deployment questions.