Overview
The most powerful tool in the Bilt MCP arsenal. Send natural language instructions to build features, deploy apps, or modify existing code.This tool triggers asynchronous workflows. Use bilt_get_session or bilt_get_messages to monitor progress.
Parameters
Active workflow session ID from bilt_get_session.
- Format: UUID
- Example:
"550e8400-e29b-41d4-a716-446655440000" - Get it: Call
bilt_get_session()first
Natural language instruction for the AI agent.
- Min length: 1 character
- Max length: 5000 characters
- Be specific: Clear instructions get better results
- Examples:
"Add a login screen with email and password fields""Deploy to production""Fix the undefined variable error in UserComponent"
Whether to wait for workflow completion before returning.
- false (default): Returns immediately with workflow ID
- true: Blocks until workflow completes (up to 5 minutes)
- Use true for: Short operations, synchronous agents
- Use false for: Long builds, async agents
Response
Session ID for this workflow
Whether the workflow was successfully initiated
Number of active SSE subscribers (usually 0 for API calls)
Unique identifier for this specific workflow run
Current workflow status:
"pending", "running", "completed", "failed"Example Usage
Basic Feature Addition
Deployment
Synchronous Operation
Agent Workflow Examples
Pattern 1: Build and Deploy
Pattern 2: Iterative Development
Message Best Practices
Be specific and clear
Be specific and clear
✅ Good:❌ Too vague:
Break complex requests into steps
Break complex requests into steps
Instead of one huge message:Break it down:
"Create product listing page""Add shopping cart functionality""Implement checkout flow""Integrate payment processing"
Include context for modifications
Include context for modifications
✅ Good:❌ Missing context:
Use deployment keywords
Use deployment keywords
Trigger deployment with clear phrases:
"Deploy to production""Publish the app""Go live""Deploy now"
Understanding Workflow States
The workflow goes through these states:pendingQueued, waiting to start
runningActively building/deploying
completedSuccessfully finished
failedError occurred, check logs
Monitoring Progress
Since workflows are async, poll for updates:Error Handling
Performance Tips
Batch related changesSend one message with multiple changes instead of multiple messages
Use async modeSet
wait_for_response: false for long buildsPoll efficientlyCheck status every 2-5 seconds, not every second
Cache session IDsReuse the same session for multiple messages
Rate Limits
Messages100 messages per hour per project
Concurrent Workflows1 workflow per session at a time
