Skip to main content

Configuring MCP Agent Actions

After creating an MCP agent, you need to configure what actions it can perform. These actions determine what capabilities your agent exposes to connected tools like Claude Desktop, Cursor, and Claude Code.

Understanding Agent Actions​

When you connect a tool to your MCP agent, the actions you configure become available as "tools" within that application. For example:

  • An action to search documentation
  • An action to query specific data types
  • An action with metadata filters for precise retrieval

Configuring Your Agent​

Step 1: Access Your MCP Agent​

  1. Navigate to Agents in your workspace
  2. Select your MCP agent
  3. Look for the configuration section where you define agent capabilities

Step 2: Define Actions​

Each action needs:

  • Name: A clear, descriptive identifier (e.g., search-docs)
  • Description: Help the AI understand when to use this action
  • Pipeline: Which data pipeline this action queries
  • Parameters (optional): Filters, search criteria, or other inputs

Step 3: Write Clear Descriptions​

Good descriptions help the AI choose the right action:

❌ Poor: "Search"
✅ Good: "Search technical documentation including API specs, architecture guides, and runbooks"

❌ Poor: "Get customer data"
✅ Good: "Query customer support tickets with filters for status, priority, date range, and product area"

Example Configurations​

Name: search-engineering-docs
Description: Search internal engineering documentation, code examples, and technical specifications
Pipeline: engineering-knowledge-base

Customer Support​

Name: query-support-tickets  
Description: Retrieve customer support tickets and feedback, filterable by status, priority, and date
Pipeline: customer-support-data

Code Analysis​

Name: analyze-codebase
Description: Search and analyze code patterns, dependencies, and architecture decisions
Pipeline: code-repository-index

Using Metadata Filters​

If your pipeline has metadata fields, you can create targeted actions:

  • Filter by document type: API docs, tutorials, references
  • Filter by date: Recently updated, specific time ranges
  • Filter by category: Backend, frontend, infrastructure
  • Filter by status: Published, draft, archived

Best Practices​

  1. Start simple: Create one general search action first
  2. Add specific actions: Create targeted actions for common queries
  3. Use clear naming: Make action names self-explanatory
  4. Test thoroughly: Verify each action works before connecting tools
  5. Iterate based on usage: Refine actions based on how they're used

Troubleshooting​

Actions not appearing in tools?

  • Ensure your MCP agent is properly configured
  • Check that your API key has the correct permissions
  • Verify the tool is connected to the right agent ID

Getting irrelevant results?

  • Refine your action descriptions
  • Add metadata filters to narrow scope
  • Consider creating more specific actions

Next Steps​

Was this page helpful?