Skip to main content
Connic
Connectors

MCP Server

Expose your agents as MCP tools. Link agents to the connector and they become callable tools via the MCP protocol.

Last updated
Looking for Connic MCP?
Need to manage a Connic project from an AI client? Use AI agent setup.
Supported MCP protocols

One endpoint supports stateless 2026-07-28 requests with server/discover; Streamable HTTP versions 2025-11-25, 2025-06-18, and 2025-03-26 with initialize and notifications/initialized; and HTTP/SSE version 2024-11-05. tools/list and tools/call work with every supported version. The 2026-07-28 discovery and tool-list responses use ttlMs: 0 and cacheScope: private, so clients revalidate them instead of sharing cached results.

Protocol scope

GET/SSE is available with 2024-11-05, and ping is available with the non-stateless protocol versions. Neither is part of the 2026-07-28 stateless protocol. Connic exposes linked agents as tools only: it does not expose MCP resources, prompts, or subscriptions, and it does not implement the optional Tasks or MCP Apps extensions.

Compare the MCP server, connector, and client roles if you also need an agent to consume tools from an external MCP server.

Sync Mode (Recommended)
Waits for the agent to complete and returns the result. Best for interactive use where you need immediate feedback.

Setup Instructions

1

Create the connector

Open your agent, click Add inbound connector in the Connector Flow, then Create New Connector and select MCP Server.

2

Choose Sync mode and copy credentials

Select Sync mode and click Create. The detail drawer opens automatically with the Endpoint URL and Secret in the Configuration section. All linked agents become callable MCP tools. To view credentials later, hover over the connector in the Connector Flow and click the Eye icon.

3

Add to your MCP client

Paste the endpoint into your MCP client config. Use 2026-07-28 stateless requests, one of the supported Streamable HTTP versions, or 2024-11-05 HTTP/SSE.

Response Format

This example shows the 2026-07-28 response. Other supported versions receive the same tool content without resultType.

response.json
{
  "jsonrpc": "2.0",
  "id": 1,
  "result": {
    "resultType": "complete",
    "content": [
      {
        "type": "text",
        "text": "The invoice total is $1,234.56. Payment due: Jan 15, 2025."
      }
    ],
    "isError": false,
    "_meta": {
      "io.modelcontextprotocol/serverInfo": {
        "name": "connic-mcp-server",
        "title": "Connic MCP Server",
        "version": "1.0.0"
      }
    }
  }
}
Timeout is 5 minutes. For longer tasks, use Inbound mode.
Configuration
  • Mode: Sync (wait for result) or Inbound (fire & forget)
  • Require Auth: Enable/disable secret key authentication (default: enabled)

When created, the connector generates a unique Endpoint URL and Secret Key.

Linked Agents as Tools

Link agents to the MCP connector from the agent detail page. Each linked agent is exposed as a tool with a standardized input schema:

tool-schema.json
// Each linked agent becomes a tool with this schema
{
  "name": "agent_name",        // Agent name (lowercase, underscores)
  "title": "Agent Name",       // Display name
  "description": "Invoke the Agent Name agent",
  "inputSchema": {
    "type": "object",
    "properties": {
      "message": {
        "type": "string",
        "description": "The message or prompt to send to the agent"
      },
      "payload": {
        "type": "object",
        "description": "Additional structured data to pass to the agent"
      }
    },
    "required": ["message"]
  }
}

Agent names are converted to lowercase with spaces and hyphens replaced by underscores:Invoice Processorinvoice_processor

Calling a Tool

Use tools/call to invoke an agent:

request.json
{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "invoice_processor",
    "arguments": {
      "message": "Process this invoice and extract the total",
      "payload": {
        "invoice_id": "INV-12345",
        "customer": "Acme Corp"
      }
    },
    "_meta": {
      "io.modelcontextprotocol/protocolVersion": "2026-07-28",
      "io.modelcontextprotocol/clientInfo": {
        "name": "your-mcp-client",
        "version": "1.0.0"
      },
      "io.modelcontextprotocol/clientCapabilities": {}
    }
  }
}

The example shows the 2026-07-28 request metadata. Clients using that version also send MCP-Protocol-Version, Mcp-Method, and Mcp-Name headers. MCP SDKs handle these fields automatically.

message (required): The prompt sent to the agent

payload (optional): Additional structured fields merged into the agent input

Authentication

Include the connector's static secret key in every request using one of these headers:

terminal
# Authorization header (recommended)
Authorization: Bearer your-secret-key

# Or X-Connic-Secret header
X-Connic-Secret: your-secret-key

The connector uses a pre-shared secret. It does not provide MCP OAuth discovery or an interactive OAuth authorization flow. Browser requests that include an Origin header are rejected; use a native or server-side MCP client.

If Require Auth is disabled, requests work without a secret. Only disable for an intentionally unauthenticated endpoint.
Inbound Mode
Returns an ordinary MCP tool result containing a run ID, then processes the agent in the background. Check run status via the dashboard or API. This is Connic fire-and-forget behavior, not the MCP Tasks extension.

Setup Instructions

1

Create the connector

Open your agent, click Add inbound connector in the Connector Flow, then Create New Connector and select MCP Server.

2

Choose Inbound mode and copy credentials

Select Inbound mode and click Create. The detail drawer opens automatically with the Endpoint URL and Secret in the Configuration section. To view credentials later, hover over the connector in the Connector Flow and click the Eye icon.

3

Connect your client

Tool calls return immediately with a run ID. Results are available in the dashboard or via an outbound connector.

Response Format

This example shows the 2026-07-28 response. Other supported versions receive the same tool content without resultType.

response.json
{
  "jsonrpc": "2.0",
  "id": 1,
  "result": {
    "resultType": "complete",
    "content": [
      {
        "type": "text",
        "text": "Agent run started with ID: 550e8400-e29b-41d4-a716-446655440000"
      }
    ],
    "isError": false,
    "_meta": {
      "io.modelcontextprotocol/serverInfo": {
        "name": "connic-mcp-server",
        "title": "Connic MCP Server",
        "version": "1.0.0"
      }
    }
  }
}
Configuration
  • Mode: Sync (wait for result) or Inbound (fire & forget)
  • Require Auth: Enable/disable secret key authentication (default: enabled)

When created, the connector generates a unique Endpoint URL and Secret Key.

Linked Agents as Tools

Link agents to the MCP connector from the agent detail page. Each linked agent is exposed as a tool with a standardized input schema:

tool-schema.json
// Each linked agent becomes a tool with this schema
{
  "name": "agent_name",        // Agent name (lowercase, underscores)
  "title": "Agent Name",       // Display name
  "description": "Invoke the Agent Name agent",
  "inputSchema": {
    "type": "object",
    "properties": {
      "message": {
        "type": "string",
        "description": "The message or prompt to send to the agent"
      },
      "payload": {
        "type": "object",
        "description": "Additional structured data to pass to the agent"
      }
    },
    "required": ["message"]
  }
}

Agent names are converted to lowercase with spaces and hyphens replaced by underscores:Invoice Processorinvoice_processor

Calling a Tool

Use tools/call to invoke an agent:

request.json
{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "invoice_processor",
    "arguments": {
      "message": "Process this invoice and extract the total",
      "payload": {
        "invoice_id": "INV-12345",
        "customer": "Acme Corp"
      }
    },
    "_meta": {
      "io.modelcontextprotocol/protocolVersion": "2026-07-28",
      "io.modelcontextprotocol/clientInfo": {
        "name": "your-mcp-client",
        "version": "1.0.0"
      },
      "io.modelcontextprotocol/clientCapabilities": {}
    }
  }
}

The example shows the 2026-07-28 request metadata. Clients using that version also send MCP-Protocol-Version, Mcp-Method, and Mcp-Name headers. MCP SDKs handle these fields automatically.

message (required): The prompt sent to the agent

payload (optional): Additional structured fields merged into the agent input

Authentication

Include the connector's static secret key in every request using one of these headers:

terminal
# Authorization header (recommended)
Authorization: Bearer your-secret-key

# Or X-Connic-Secret header
X-Connic-Secret: your-secret-key

The connector uses a pre-shared secret. It does not provide MCP OAuth discovery or an interactive OAuth authorization flow. Browser requests that include an Origin header are rejected; use a native or server-side MCP client.

If Require Auth is disabled, requests work without a secret. Only disable for an intentionally unauthenticated endpoint.