mcpField
T
Tasker MCP

Tasker MCP

Local
github.com/dceluisProductivity52Updated 85w ago

About Tasker MCP

Tasker MCP is an MCP server that connects the Android automation app Tasker to MCP clients. After importing a Tasker profile, a command-line server exposes Tasker tasks as tools. It suits users who want an AI assistant to trigger phone automations.

Key features

  1. 01Enabled by importing a Tasker profile
  2. 02SSE and stdio transport modes
  3. 03API key protection for access
  4. 04Tool descriptions generated from task comments and variables
  5. 05Prebuilt binaries for several architectures

Use cases

  1. 01Trigger Tasker tasks on a phone from an AI assistant
  2. 02Expose custom Tasker tasks as MCP tools
  3. 03Connect phone automation to Claude Desktop over stdio

Configuration

Local server config is the same for every client. Copy it into your client’s MCP config file.

No install method listed yet

We haven’t collected a package or remote endpoint for this server. Follow the instructions in the project README.

Repository

Tools

No tool list yet

Local servers expose tools only once running on your machine, and the maintainer has not provided a tool list.

Details

From the project README

Tasker MCP

This document will guide you through setting up and running the Tasker MCP integration, including instructions for installing dependencies, preparing servers, and updating tasks.


Usage Guide

Step 1: Import the Tasker Profile
  • Import dist/mcp_server.prj.xml into your Tasker app.
  • After importing, run the MCP generate_api_key task to generate an API key for secure access.
Step 2: Select and Run Your Server

CLI Server:

  • From the dist/ folder, select the correct CLI server binary for your device's architecture, such as tasker-mcp-server-cli-aarch64.
  • Copy both the binary and the toolDescriptions.json file to your device (phone or PC).
  • Rename the binary to mcp-server after copying.

Example:

Using scp:

scp dist/tasker-mcp-server-cli-aarch64 user@phone_ip:/data/data/com.termux/files/home/mcp-server

Using adb push:

adb push dist/tasker-mcp-server-cli-aarch64 /data/data/com.termux/files/home/mcp-server
  • Run the server in SSE mode with:
./mcp-server --tools /path/to/toolDescriptions.json --tasker-api-key=tk_... --mode sse
  • Or call it through the stdio transport:
payload='{"jsonrpc": "2.0", "id": 1, "method": "tools/call", "params": { "name": "tasker_flash_text", "arguments": { "text": "Hi" }  } }'
echo $payload | ./mcp-server --tools /path/to/toolDescriptions.json --tasker-api-key=tk_...
Command-Line Flags

The tasker-mcp-server-cli application accepts the following flags:

  • --tools: Path to JSON file with Tasker tool definitions.
  • --host: Host address to listen on for SSE server (default: 0.0.0.0).
  • --port: Port to listen on for SSE server (default: 8000).
  • --mode: Transport mode: sse, or stdio (default: stdio).
  • --tasker-host: Tasker server host (default: 0.0.0.0).
  • --tasker-port: Tasker server port (default: 1821).
  • --tasker-api-key: The Tasker API Key.
Step 3: Connect Your MCP-enabled App
  • Connect your MCP-enabled application by pointing it to the running server.
Example Configuration for Claude Desktop with stdio transport
{
  "mcpServers": {
    "tasker": {
      "command": "/home/luis/tasker-mcp/dist/tasker-mcp-server-cli-x86_64",
      "args": [
        "--tools",
        "/home/luis/tasker-mcp/dist/toolDescriptions.json",
        "--tasker-host",
        "192.168.1.123",
        "--tasker-api-key",
        "tk_...",
        "--mode",
        "stdio"
      ]
    }
  }
}

Building the CLI Server Yourself

Unix/Linux:
  • Install Go using your package manager:
sudo apt-get install golang-go
  • Build the CLI server (cross-compiling example for ARM64):
cd cli
GOOS=linux GOARCH=arm64 go build -o dist/tasker-mcp-server-cli-aarch64 main.go

Updating the MCP Profile with Additional Tasks

Due to limitations in Tasker's argument handling, follow these steps carefully to mark tasks as MCP-enabled:

Step 1: Set Task Comment
  • Add a comment directly in the task settings. This comment becomes the tool description.
Step 2: Configure Tool Arguments Using Task Variables

Tasker supports only two positional arguments (par1, par2). To work around this, we'll use Task Variables:

  • A TaskVariable becomes an MCP argument if:
    1. Configure on Import: unchecked
    2. Immutable: true
    3. Value: empty

After setting the above values you can also set some additional metadata:

  • Metadata mapping:
    • Type: Derived from Task Variable's type (number, string, onoff, etc).
    • Description: Set via the variable's Prompt field.
    • Required: If the Same as Value field is checked.

Note: Temporarily enable "Configure on Import" to set the Prompt description if hidden, then disable it again. The prompt will survive.\

These steps will make sure valid tool descriptions can be generated when we export our custom project later.
Task Variables cannot be pass-through from other tasks, though, so we need to do one last thing in order to get all the variables from the MCP request properly set.

Step 3: Copy the special action

Copy the action MCP#parse_args to the top of your MCP task to enable argument parsing. You can get this from any of the default tasks. But do not modify this action!

Step 4: Exporting and Generating Updated Tool Descriptions

Now your custom tasks are ready:

  • Export your mcp-server project and save it on your PC.
  • Ensure Node.js is installed, then run:
cd utils
npm install
node xml-to-tools.js /path/to/your/exported/mcp_server.prj.xml > toolDescriptions.json

Use this toolDescriptions.json file with your server.


Happy automation!

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.