MCP Server

Your VAPI agent needs to look up an order. Your Retell agent needs to remember a caller from last week. Your Claude Desktop assistant needs to search your company's docs. Each of these requires a backend service you'd normally spend weeks building.

Chanl gives every workspace an MCP endpoint that handles all of it:

https://{workspace-slug}.chanl.dev/{toolset-slug}

Connect any MCP-compatible client — and your agent instantly gets tools, knowledge search, persistent memory, and prompt templates. No backend code. No infrastructure. No RAG pipeline to build.

What Your Agent Gets

One MCP connection, five backend services:

Quick Start

{
  "mcpServers": {
    "chanl": {
      "type": "streamablehttp",
      "url": "https://acme.chanl.dev/admin-tools",
      "headers": {
        "X-API-Key": "chanl_key_abc123..."
      }
    }
  }
}

{
  "mcpServers": {
    "chanl": {
      "type": "streamablehttp",
      "url": "https://acme.chanl.dev/admin-tools",
      "headers": {
        "X-API-Key": "chanl_key_abc123..."
      }
    }
  }
}

{
  "mcpServers": {
    "chanl": {
      "type": "streamablehttp",
      "url": "https://acme.chanl.dev/admin-tools",
      "headers": {
        "X-API-Key": "chanl_key_abc123..."
      }
    }
  }
}

import { Client } from '@modelcontextprotocol/sdk/client';
import { StreamableHTTPClientTransport } from '@modelcontextprotocol/sdk/client/streamableHttp.js';

const transport = new StreamableHTTPClientTransport(
  new URL('https://acme.chanl.dev/admin-tools'),
  {
    requestInit: {
      headers: { 'X-API-Key': process.env.CHANL_API_KEY! },
    },
  },
);

const client = new Client({ name: 'my-app', version: '1.0.0' });
await client.connect(transport);

// Search your knowledge base during a call
const result = await client.callTool({
  name: 'kb_search',
  arguments: { query: 'refund policy for orders over 30 days', limit: 5 },
});
console.log(result.content); // Relevant policy documents

// Remember something about this customer
await client.callTool({
  name: 'memory_add',
  arguments: {
    customerId: 'cust_123',
    content: 'Prefers email follow-ups. Has 2 active subscriptions.',
  },
});

await client.close();

MCP primitives: tools, resources, prompts

The Model Context Protocol defines three discovery surfaces. Chanl implements all three on every workspace MCP endpoint:

Workspace prompts (from your library)

Prompts you create via the Prompts API or SDK are registered on the MCP server automatically. Each prompt becomes an MCP prompt with variables as arguments. Clients call prompts/get to render Liquid templates at runtime.

# Manage the library
chanl prompts list
chanl prompts create -f greeting.json

Guided prompts (shipped with Chanl)

These static MCP prompts are always available, independent of your workspace library:

• create-agent — interview workflow for a new agent
• review-customer — 360° customer review
• triage-followups — prioritize follow-ups
• build-tool — design an HTTP or system tool
• create-scenario — draft a test scenario
• improve-agent — review and improve agent instructions

Prompt tools (tool-calling clients)

Some clients only support tools, not MCP prompts. Chanl also exposes prompt_list and prompt_get as admin tools that call the same REST API. Prefer MCP prompts when your client supports them — they are the protocol-native surface.

Built-in Tools

Every workspace gets an admin-tools toolset at slug admin-tools. No setup needed — these are available the moment you create a workspace.

Custom Toolsets

The built-in admin-tools are a starting point. For production, you'll want toolsets scoped to specific agents or use cases.

A customer support agent doesn't need workspace_stats. It needs lookup_order, check_warranty, and create_ticket. Here's how to set that up:

# Create a scoped toolset
chanl toolsets create --name "Support Tools" --slug "support-tools"

# Add an HTTP tool that calls your order API
chanl tools create --name "lookup_order" --type http \
  --method GET --url "https://api.yourshop.com/orders/{{orderId}}" \
  --headers '{"Authorization": "Bearer {{secret:shop_api_key}}"}'

# Add it to the toolset
chanl toolsets add-tool --toolset support-tools --tool <tool-id>

Now your support agent connects to https://acme.chanl.dev/support-tools and gets exactly the tools it needs. Nothing more, nothing less.

How MCP Fits In

The conversation happens on your platform. The intelligence comes from Chanl. And every call is automatically transcribed, scored, and tracked — whether it's a test simulation or a production call.

Authentication & URLs

All requests require X-API-Key header. Keys are workspace-scoped — acme.chanl.dev only accepts keys from the acme workspace.

Troubleshooting

What's Next