Skip to main content
Create, generate, manage, and govern MCP Foundry capabilities. Use these workflows to turn enterprise systems and knowledge into reliable agent capabilities. Each workflow ends by attaching the result to a virtual server and testing it before broader use.

​
Generate tools from API definitions

The MCP Tools Generator accelerates API onboarding. It accepts OpenAPI specification files or URLs, API documentation, and Postman collections, then produces candidate tool definitions that should be reviewed before publication.
  1. Open MCP Tools Generator.
  2. Choose the source type and provide the file or URL.
  3. Start the generation job and follow its status in Generation Jobs.
  4. Review names, descriptions, endpoints, request schemas, authentication assumptions, and visibility for each generated tool.
  5. Test generated tools with non-production credentials.
  6. Attach only approved tools to a virtual server.
Caution: Generation is an onboarding aid, not an authorization decision. Check every generated endpoint and credential scope, especially for write or destructive operations.

​
Turn knowledge into agent-ready resources

Resources supply the context agents need to make informed decisions.
  1. Open Resources and choose Add Resource.
  2. Add an approved URL, or paste inline content.
  3. Set a clear name, description, resource type, tags, owner, and visibility.
  4. Save the resource and attach it to the relevant virtual server.
  5. Use Playground to verify the agent can read the intended content and cannot access content outside the server’s composition.
Use private visibility for narrowly authorized use, internal for organization-wide availability under your policy, and public only when the content is deliberately safe for broad compatible-client access.

​
Create a reusable prompt

Prompts separate reusable instructions from application code.
  1. Open Prompts and select Add Prompt.
  2. Provide a descriptive name, description, and tags.
  3. Write the template using Jinja-style variables and logic.
  4. Declare each argument the template expects.
  5. Set visibility and ownership, then save.
  6. Attach the prompt to a virtual server and test it with representative values.
Keep templates explicit about role, permitted tools, required context, refusal or escalation conditions, and output format. Prompt arguments should make context requirements visible instead of relying on hidden client-side assumptions.

​
Build a portable skill

A skill is a reusable operating procedure that agents can consume across supported clients and gateways.
  1. Open Skills Hub and choose New Skill.
  2. Add a name, description, and optional compatibility or licensing information.
  3. Select the tools the skill is allowed to use.
  4. Choose which MCP gateways may expose the skill, where applicable.
  5. Set visibility and ownership.
  6. Write the markdown instructions: objective, prerequisites, procedure, guardrails, escalation rules, and response format.
  7. Create the skill, preview or download its markdown as needed, then attach it to a virtual server.
A good skill is outcome-focused. It tells an agent how to perform a task consistently; it does not simply list available tools.

​
Connect and publish MCP servers

Use MCP Servers to manage external, self-hosted, or deployable MCP services. Browse the registry for pre-integrated options, or register a server using the connection approach appropriate for the deployment. For command-line servers, MCP Foundry can bridge the server so agents can reach it through a managed network endpoint. After connecting a server:
  1. Verify its health and capability catalog.
  2. Review the imported tools, resources, and prompts.
  3. Restrict access through teams, visibility, and virtual-server composition.
  4. Test a representative call in Playground.
  5. Monitor logs and error rates after release.

​
Govern access and operations

  • Use Teams to establish ownership and scoped collaboration.
  • Use API Tokens for programmatic access; rotate and revoke them according to your security policy.
  • Configure Roots to constrain filesystem access for server workloads to explicitly permitted directories.
  • Use Logs to trace activity and investigate failures.
  • Use dashboard metrics to identify slow endpoints, error-prone tools, and unusual traffic patterns.
For a complete first composition, continue with Create your first MCP server.