MCP Integration

ApiMeld exposes a Model Context Protocol (MCP) server, allowing AI assistants like Claude to interact with your tasks, data sources, execution history, and notification logs directly — without leaving the conversation.

What MCP lets you do

Once connected, an AI assistant can:

• List, create, update, and run tasks
• Toggle task schedules on or off
• Check execution status and stream logs after a run
• Browse available data sources and their endpoints
• Query notification delivery history

All operations respect the same permission model as the UI — the AI acts as the user whose API key was used to connect.

Prerequisites

• An ApiMeld API key (see API Keys)
• An MCP-compatible AI client (Claude Code, Claude Desktop, or any client supporting the MCP HTTP transport)

Setting up the connection

Claude Code (CLI)

Run once to register the server permanently:

claude mcp add apimeld --transport http https://your-apimeld-host/mcp \
  --header "Authorization: Bearer msk_your_key_here"

The configuration is saved to ~/.claude.json and loaded automatically in every future session. The MCP tools appear in the session as soon as Claude Code starts — no further setup needed.

To use the internal address instead (e.g. on the same network):

claude mcp add apimeld --transport http http://192.168.0.33:8080/mcp \
  --header "Authorization: Bearer msk_your_key_here"

To verify the connection, ask Claude to list your tasks:

"List all my scheduled tasks"

Claude Desktop

Add the following to your claude_desktop_config.json:

{
  "mcpServers": {
    "apimeld": {
      "type": "http",
      "url": "https://your-apimeld-host/mcp",
      "headers": {
        "Authorization": "Bearer msk_your_key_here"
      }
    }
  }
}

Config file locations:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

Restart Claude Desktop after saving.

Other MCP clients

The ApiMeld MCP endpoint is a standard Streamable HTTP transport at:

POST https://your-apimeld-host/mcp
Authorization: Bearer msk_your_key_here

Any client that supports the MCP HTTP transport can connect using those details.

Available tools

Tool	Description
list_tasks	List all tasks visible to you
get_task	Full task detail including script body
create_task	Create a new task
update_task	Partially update a task
delete_task	Delete a task
run_task	Trigger an immediate run, returns an executionId
toggle_task	Pause or resume a task's schedule
get_execution_status	Poll run status and logs by executionId
list_task_history	Last N runs for a task
list_datasources	Data sources accessible to you
get_datasource_endpoints	OpenAPI endpoints for a REST data source
list_notification_logs	Notification delivery history with filtering

Typical workflow

You: Run the "Daily Report" task and tell me if it succeeded.

Claude: [calls run_task] → executionId 1042
        [polls get_execution_status] → Success, 3.2s
        The task ran successfully in 3.2 seconds with no errors.

You: Did the last MQTT temperature alert send to Slack?

Claude: [calls list_notification_logs with taskId + channel=Slack]
        Yes — sent to the "Alerts" channel at 10:20 PM.
        Body: "Temperature alert exceeded: 105.5"

Security notes

• The AI inherits exactly your user's permissions — it cannot access tasks or data sources you cannot access
• API keys used for MCP follow the same restrictions as all API keys: admin config changes and key management require interactive login (see API Keys)
• Create a dedicated API key for your AI assistant so you can revoke it independently if needed
• The MCP endpoint requires authentication — unauthenticated requests are rejected