MCP Server · 38 tools · Full lifecycle

Deploy AI Agents
from your terminal

KILN is an MCP server. Build, deploy, and manage agents from Claude Code, Cursor, or any MCP client. 38 tools. Full lifecycle.

Get your API key — freeSee the setup
Claude Code Cursor Windsurf Any MCP Client

Quick Start

Get started in 30 seconds

1

Add the MCP server

Terminal
$ claude mcp add kiln-mcp \
  --transport http \
  https://kilnbase.com/api/mcp \
  -H "Authorization: Bearer sk-kiln-YOUR_KEY"
2

Or configure via JSON

.cursor/mcp.json or claude_desktop_config.json
{
  "mcpServers": {
    "kiln-mcp": {
      "url": "https://kilnbase.com/api/mcp",
      "headers": {
        "Authorization": "Bearer sk-kiln-YOUR_KEY"
      }
    }
  }
}
3

Start building

Just tell your AI assistant what to build
You: "Create a support agent for acme.com that answers questions about our product, collects leads, and escalates complex issues."
KILN: Agent created — "Acme Support" Status: LIVE URL: kilnbase.com/a/acme-support Actions: COLLECT_EMAIL, SCORE_LEAD, HANDOFF_HUMAN Knowledge: Crawling acme.com... 12 pages indexed

Capabilities

What you can do

Everything from agent creation to production monitoring, all from your IDE.

Create agents

Describe what you need in natural language. KILN generates the system prompt, configures actions, and deploys instantly.

"Create a booking agent for my dental practice"

Train on data

Add URLs, PDFs, FAQs, and text to the knowledge base via MCP. Auto-chunking and vector embeddings.

kiln_add_knowledge({ type: "URL", sourceUrl: "docs.acme.com" })

Deploy anywhere

Embed widget on any site, connect Slack, WhatsApp, Telegram, Email, or GitHub. One agent, every channel.

kiln_get_embed_code({ agentId }) → <script>

Build teams

Multi-agent orchestration from your terminal. Create teams with HEAD, COORDINATOR, and EXECUTOR roles.

kiln_create_team({ template: "SALES" })

Monitor quality

Create test suites, run evals, analyze knowledge gaps, and A/B test different configs programmatically.

kiln_run_tests({ agentId }) → { score: 0.92 }

Scale with API

Scoped API keys, inbound webhooks, scheduled automations, and workflow orchestration for production workloads.

kiln_create_workflow_automation({ schedule: "0 9 * * 1" })

Infrastructure as Code

Version control
your AI agents

Export agent configs as JSON, store them in Git, and import them anywhere. Treat AI agents like any other infrastructure.

Start building
agent-config.json
{
  "name": "Acme Support",
  "systemPrompt": "Du bist der Kundenservice...",
  "llmModel": "claude-sonnet-4-6",
  "temperature": 0.7,
  "actions": [
    { "type": "COLLECT_EMAIL", "enabled": true },
    { "type": "SCORE_LEAD", "enabled": true },
    { "type": "BOOK_APPOINTMENT", "config": {
      "calendlyUrl": "https://cal.com/acme"
    } }
  ],
  "whiteLabel": {
    "primaryColor": "#2563EB",
    "showPoweredBy": false
  }
}

Reference

Full tool reference

38 MCP tools across 8 categories. Click to expand.

kiln_list_agentsList all agents with status, model, conversation count, and public URL.
kiln_create_agentCreate a CHAT or TASK agent from natural language. Returns ID, slug, and URL.
kiln_update_agentUpdate any agent config field — prompt, model, personality, language.
kiln_delete_agentPermanently delete an agent and all associated data.
kiln_clone_agentClone an agent with all config, actions, tools, and knowledge.
kiln_deploy_agentSet agent status to LIVE, DRAFT, or PAUSED.
kiln_export_agent_configExport full agent config as JSON for version control.
kiln_import_agent_configCreate an agent from exported config JSON.

Computer Use

Computer Use &
Sandbox Execution

Give your agents real computer capabilities. Browse websites, fill forms, extract data, and execute code in isolated sandboxes — all orchestrated through KILN.

Browser Automation

Agents navigate real websites, click buttons, fill forms, and extract structured data. Powered by Playwright with anti-detection.

Visual Understanding

Screenshot-based reasoning. Agents see the page like a human, identify elements, and take action based on visual context.

Code Sandbox

Execute Python, JavaScript, and TypeScript in isolated E2B sandboxes. Install packages, process files, run data pipelines.

Isolated Execution

Every sandbox is a fresh VM with resource limits. No cross-session contamination. Auto-cleanup after execution completes.

Persistent Workspace

Agents save files, reports, and artifacts to persistent workspaces. Version history, download API, and plan-gated storage limits.

Checkpoint & Resume

Long-running browser and code tasks checkpoint automatically. If the serverless function times out, execution resumes from the last checkpoint.

Learn more about Computer Use

Agent SDK

Agent SDK & CLI

Build, deploy, and manage agents programmatically with our TypeScript SDK and CLI tool.

TypeScript SDK

First-class TypeScript support with full type definitions. Create agents, manage knowledge bases, and handle conversations from your codebase.

CLI Tool

Deploy agents from CI/CD pipelines, export configs, run tests, and manage infrastructure — all from the command line.

RBAC Scoped Keys

Generate API keys with granular permissions. Restrict access to specific agents, actions, or read-only analytics.

Extensibility

Custom Node SDK

Build custom workflow nodes with our Node SDK. Define inputs, outputs, and execution logic — then share them on the Marketplace.

Define Schema

Strongly typed input/output schemas with validation. Your nodes appear in the visual workflow editor with auto-generated UI.

Sandboxed Execution

Nodes run in isolated sandboxes with resource limits. Access HTTP, file system (scoped), and environment variables securely.

Marketplace Publishing

Package and publish custom nodes to the KILN Marketplace. Version control, documentation, and usage analytics included.

Pre-built Connectors

Start from 20+ pre-built connectors (Slack, Gmail, Sheets, Notion) or fork them to build your own.

Embed

Embeddable Components

Drop KILN agents into any frontend. React, Vue, vanilla JS — or just an iframe.

React SDK

<KilnChat /> component with full props API. Theme customization, event hooks, and TypeScript support.

npm install @kiln/react

Web Component

Framework-agnostic <kiln-chat> custom element. Works with Vue, Svelte, Angular, or plain HTML.

<script src="kiln.js"></script>

iframe Embed

Zero-dependency embed with one line. Full white-label support, responsive sizing, and cross-origin messaging.

<iframe src="/embed/slug" />

Execution

Sandbox API

Execute arbitrary code in isolated sandboxes. Perfect for AI-generated code, data transformations, and custom agent logic.

Language Support

Run JavaScript, TypeScript, and Python in sandboxed environments with pre-installed packages and dependency management.

Resource Limits

Configurable CPU, memory, and execution time limits. Auto-termination for runaway processes. Scoped filesystem access.

REST API

Simple POST endpoint. Send code, get results. Supports stdin/stdout streaming for long-running tasks.

Integrations

Webhook V2

Production-grade webhook infrastructure with retry logic, HMAC signatures, and delivery tracking.

HMAC Signatures

Every payload is signed with SHA-256 HMAC. Verify authenticity on your end with the shared secret.

Auto Retry

Failed deliveries are retried with exponential backoff. Up to 5 attempts over 24 hours.

Delivery Logs

Full history of every webhook delivery — payload, response code, latency, and retry count.

Event Filtering

Subscribe to specific events: conversation.started, lead.captured, agent.deployed, task.completed.

Ready to build?

Get your free API key and start deploying agents in minutes.

Get your API key — free

Free tier includes 50 AI credits/month. No credit card required.