Skip to main content
This quick start creates a small, governed MCP surface: one tool, one resource, and one virtual server. It is designed to establish the configuration pattern before you connect production systems.

Before you begin

You need:
  • Access to the MCP Foundry dashboard.
  • Permission to create tools, resources, and virtual servers in the target team.
  • A safe HTTP API endpoint for the example tool, or an existing approved tool.
  • An MCP-compatible client or the built-in Playground to test the result.
Tip: Use a non-production endpoint and non-sensitive content for your first server. Verify authorization and input schemas before exposing a production API to agents.

1. Create a tool

  1. In the dashboard, open Tools and select Add Tool.
  2. Enter a unique name, a human-readable display name, description, and optional tags.
  3. Configure the HTTP endpoint, method, required headers, and the authentication settings for the backend API.
  4. Define a JSON Schema for the arguments the agent may send. Make required fields explicit and keep the schema as narrow as practical.
  5. Select the appropriate visibility and save the tool.
A useful first schema has a single required lookup argument:

2. Add a resource

  1. Open Resources and select Add Resource.
  2. Provide a resource URL, or enter inline content when the material is not hosted elsewhere.
  3. Add a name, description, resource type, tags, ownership, and visibility.
  4. Save the resource.
For a first test, add a short support-policy summary. Resources are context for agents; they do not automatically grant access to an API or bypass a tool’s authorization requirements.

3. Create a virtual server

  1. Open Virtual Servers and select Create Server.
  2. Give the server a clear name and description, such as support-assistant-sandbox.
  3. Configure how MCP clients connect and choose a visibility appropriate to the environment.
  4. Attach the tool and resource created in the previous steps.
  5. Optionally attach an existing prompt or skill.
  6. Create or publish the server.

4. Test the published server

  1. Open Playground.
  2. Select the virtual server from the MCP server list.
  3. Confirm the expected tool, resource, prompt, and skill entries are available.
  4. Invoke the tool with a permitted test value and confirm the result.
  5. Read the resource and verify that only the intended content is returned.
  6. Review Logs and the dashboard for the request outcome, latency, and errors.
If you use another MCP client, connect it to the endpoint and use the standard MCP discovery methods before calling capabilities:

What to do next

  • Add task instructions by creating a prompt or skill.
  • Connect a third-party or self-hosted server through the MCP Servers area.
  • Review concepts to plan visibility, teams, and composition boundaries.
  • See troubleshooting if the server is not discoverable or a call fails.