Local tools for FTC robot code v0.2

Your AI can
build the robot now.

FTC Toolchain gives Codex, Claude, and other MCP clients the tools to scaffold subsystems, wire TeleOp, build with Gradle, and deploy to your Control Hub.

Get started ↓Read the docs ↗

✓ Official FTC SDK ✓ Pedro Pathing ✓ Free & open source

codex — ~/DecodeRobotMCP CONNECTED
›

Make an intake subsystem with one motor, spinIn and spinOut, plus a bench test.

●

I’ll scaffold the subsystem, wire its hardware config, then run a build.

⚡ftc-toolchain.create_subsystem
DONE

Created subsystems/Intake.java · 48 lines

⚡ftc-toolchain.create_test_opmode
DONE

Created opmodes/IntakeBenchTest.java

✓BUILD SUCCESSFULin 4sAPK READY

START BUILDING

Connect your existing
robot project.

Choose your AI client, enter the FTC SDK project root—the folder containing TeamCode—then copy the generated setup.

1. Choose how to begin
2. Choose your client
3. Choose your computer
Before you startNode 18+GitAndroid Studio + JDK 17adb for robot deploys
1. Register FTC ToolchainTerminal
codex mcp add --env FTC_TOOLCHAIN_PROJECT_DIR="/Users/team/DecodeRobot" ftc-toolchain -- npx -y ftc-toolchain
2. Download FTC + Pedro referencesTerminal · run once
npx -y ftc-toolchain setup
3
Start a new AI session

Ask: “Run inspect_project before changing anything.” It should report the selected path, OpModes, Git state, Gradle, Android SDK, and hardware names.

THE GAP

Chat is useful.
Shipping robot code is better.

Every season starts with hardware maps, drive code, and config names that must match the Driver Station.

Chatbots can describe the code. You still copy, paste, fix imports, and chase build errors.

FTC Toolchain closes the loop. The AI runs real tools, builds the code, and tells you exactly what broke.

35 TOOLS. FOUR JOBS.

Everything between
the prompt and the robot.

Purpose-built tools for the repetitive, fragile parts of an FTC codebase. Your AI handles the loop; your team reviews every diff.

01↗

Knowledge

Real APIs. Real examples.

Searches official FTC samples and Pedro Pathing docs, so your AI cites working code instead of inventing methods.

search_knowledge
02↗

Scaffold

Clean code, wired up.

Creates one class per mechanism, injects dependencies, and keeps driver bindings in a separate, editable file.

create_subsystem
03↗

Build + deploy

Cable or Wi-Fi. Your choice.

Deploys directly over USB-C, or switches to saved Control Hub Wi-Fi and restores your internet automatically.

deploy_robot
04↗

Robot memory

Next session starts ahead.

Maintains a living docs/ knowledge base of your robot, hardware names, and public subsystem commands.

document_robot

THE LOOP

You describe the robot.
The tools do the typing.

Nothing is hidden. You stay in control, review the code, and decide when it touches hardware.

✦
01

Describe

Say what the mechanism should do.

→
⌘
02

Scaffold

Review a clean, FTC-native implementation.

→
>_
03

Build

Gradle runs. Compiler errors come back.

→
⇧
04

Deploy

Send it to the Control Hub when you’re ready.

PROOF, NOT A DEMO

We rebuilt a real competition robot from a prompt.

Eight subsystems. Dual TeleOps. A color-sorting spindexer with custom PID. We started with an empty folder and a plain-English description.

The AI wrote the logic. FTC Toolchain guaranteed the structure, wiring, and a buildable result.

8/8

subsystems reproduced

76/76

public methods matched

01

real, installable APK

Controls.javaGENERATED · HUMAN-EDITABLE
// Driver bindings stay out of robot logic
public void bind(GamepadEx driver) {
  driver.getGamepadButton(A)
    .whenPressed(intake::spinIn)
    .whenReleased(intake::stop);
}
✓

YOU’RE IN CONTROL

Review the diff.
Then run the robot.

FTC Toolchain proposes and runs local tools. It won’t overwrite existing files by default, and nothing deploys to your Control Hub unless you ask.

FAQ

The practical questions.

Do I need to understand MCP?+

No. Add the server with one command, run setup, then talk to your AI like you normally would.

Does it require Pedro Pathing?+

No. Pedro is optional, and FTC Toolchain includes a tool to install it when your robot needs it.

Will it overwrite my code?+

Not silently. File-generating tools refuse to overwrite existing work unless you explicitly allow it.

Does the robot need to be connected?+

Only for deployment and device logs. Scaffolding, docs, and local builds work without a robot.

How does deployment connect?+

Choose direct USB-C when you are near the robot, or automatic saved-Wi-Fi switching when you want to stay cable-free. No phone tether is required.

Where does my code go?+

The MCP server runs locally. Your AI client’s normal privacy and data settings still apply.