Skip to main content
This guide walks you through the complete Server Foundry workflow — from providing your API source to deploying a production-ready MCP server.

​
Prerequisites

  • An API source: OpenAPI spec (JSON or YAML), Postman Collection, or API documentation URL
  • A Caylex account with access to the Server Foundry

​
Phase 1: Provide your API source

1

Open Server Foundry

Navigate to Server Foundry in the Caylex dashboard. You are presented with the Designer Agent workspace.
2

Choose your input method

Select how you want to provide your API:
  • Upload an OpenAPI spec — drag and drop a JSON or YAML file. Caylex extracts endpoints, schemas, and server URLs.
  • Upload a Postman Collection — drag and drop your collection file. Caylex converts requests into tool definitions.
  • Provide an API documentation URL — enter the URL of the API’s documentation site. Caylex crawls the pages and extracts endpoints. You review and approve all discovered pages before proceeding.
3

Configure server URL (if needed)

If your spec defines multiple server URLs or uses URL variables, you are prompted to select and configure them:
  • Select a server URL — choose the base URL for your API
  • Set variables — provide values for any URL template variables (e.g., {region}, {version})
4

Start the conversation

The Designer Agent greets you and begins analyzing your API. It has full context on the extracted endpoints, schemas, and structure.

​
Phase 2: Use case discovery

The Designer Agent analyzes your API and identifies the key use cases it supports.
1

Review extracted use cases

The agent presents a list of use cases it identified from your spec. These appear in the workspace panel on the right. Examples:
  • “Manage customer records (create, update, search)”
  • “Track orders and fulfillment status”
  • “Generate and download reports”
2

Confirm or refine

Chat with the agent to refine the use cases. You can:
  • Confirm the use cases are correct
  • Ask to add missing use cases
  • Ask to remove irrelevant ones
  • Provide additional context about how your API is used
The more context you give the agent about how your API is actually used, the better it can group endpoints into useful templates in the next phase.

​
Phase 3: Template generation

Once use cases are confirmed, the agent generates workflow templates — logical groupings of endpoints that support each use case.
1

Wait for generation

Template generation takes a moment depending on the size of your spec. A progress indicator shows the estimated time.
2

Review templates

The workspace switches to the Template Selector view. Templates are organized as groups of related tools (e.g., “Customer Management” might include list_customers, get_customer, create_customer, update_customer).

​
Phase 4: Tool selection

Choose which tools to include in your generated server.
1

Browse templates and tools

The Template Selector shows two views:
  • Templates tab — workflow groups with checkboxes to select entire groups
  • All Tools tab — every available tool with individual checkboxes
You can search and filter tools by name or HTTP method.
2

Select your tools

Check the tools you want to include. The right panel shows a preview of your selected tools.
Start with a focused set of tools for your primary use case. You can always generate additional servers for other use cases later.
3

Save your selection

Click Save when you are satisfied with your tool selection. The right panel shows a summary of your selected tools.

​
Phase 5: Toolset validation

After you finalize your tool selection, Caylex validates that the selected tools can be sequenced into usable workflows. This step maps the input/output dependencies across all endpoints to ensure your agent can actually use the tools together. The validation checks for:
  • Parameter usability issues — endpoints that require IDs or references but no corresponding endpoint to discover them
  • Missing dependencies — tools that depend on output from tools not included in your selection
  • Workflow completeness — whether the selected tools form coherent end-to-end workflows
Review any validation warnings and adjust your selection if needed. This ensures your generated server produces a toolset that agents can reliably work with.

​
Phase 6: Tool optimization

The Designer Agent optimizes your tool names and descriptions for AI agent consumption.
1

Request optimization

Click Ready to proceed? or tell the agent you are ready. The agent analyzes your selected tools and applies best practices for AI tool definitions.
2

Review optimizations

The workspace switches to the Tool Optimizer view. For each tool, you see:
  • Original name and optimized name — e.g., POST /api/customers becomes customers_create
  • Original description and optimized description — enhanced with context, parameter guidance, and safety warnings
  • Optimization rationale — why the name or description was changed
  • Best practices applied — which principles guided the optimization
Optimizations follow industry best practices for AI tool naming:
  • Namespace by service and resource (e.g., cal_booking_create)
  • Context-rich descriptions with agent-specific guidance
  • Token-efficient names
  • Safety warnings for destructive operations (delete, update)
3

Accept, reject, or edit

For each tool, you can:
  • Accept — use the optimized name and description
  • Reject — keep the original name and description
  • Edit — modify the optimized values to your preference
Use Accept All or Reject All for bulk actions.Click Save when you are satisfied with the optimizations.

​
Phase 7: Server preview and generation

Configure your server and deploy it.
1

Configure server settings

The workspace switches to the Server Preview view. Configure:
  • Server name (required) — a descriptive name (3-50 characters, alphanumeric with spaces and hyphens)
  • Description (optional) — what this server does
  • Auth level — User-level or Project-level authentication
  • Server URL — confirm or adjust the base URL and variables
  • Auth headers — configure any authentication headers the server requires
2

Review the final tool list

The preview shows the complete list of tools that will be generated, with their final names (optimized if accepted), endpoints, and schemas.
3

Generate the server

Click Generate Server. Caylex:
  1. Generates the MCP server code
  2. Deploys the server to Caylex’s infrastructure
  3. Registers all tools in the platform
  4. Creates health check monitoring
Your server is generated and appears in the Servers list. You can now add it as a Server Instance to any project.

​
After generation

Once your server is generated:
  1. Add to a project — go to your project and add the new server as a Server Instance
  2. Set up authentication — create an Auth Link for users to provide their credentials (if the server requires authentication)
  3. Configure navigator access — ensure your Navigator Instance has access to the new server’s tools
  4. Test in the playground — use the Caylex Playground to verify the tools work as expected

​
Workspace management

Server Foundry automatically saves your progress. You can:
  • Resume a session — open the Saved Workspaces drawer and click on a previous session to continue where you left off
  • Start fresh — click New Workspace to begin a new server generation
  • Delete a workspace — remove workspaces you no longer need
Changes to earlier phases (like regenerating templates) automatically invalidate later phases (selection, optimization, preview). This ensures your final server always reflects your latest choices.