How to Use Jcode?

Once jcode is installed, using it comes down to five things: knowing the core commands, connecting an AI provider, optionally running a local model instead of a paid one, adding tools through MCP servers, and turning on browser automation if your work needs it.

Quick Overview

Here are the commands you’ll use most often once jcode is set up.

jcode
jcode run "say hello"
jcode --resume fox
jcode serve
jcode connect
jcode dictate

The first command opens the interactive interface. The rest let you run a single task without opening the full interface, resume a named session, run jcode as a background server, connect additional clients to that server, and send voice input from your configured speech to text tool.

jcode Command Cheat Sheet

A quick reference for the commands you’ll reach for most.

Command	What It Does
jcode	Launches the main interactive interface
jcode run "task"	Runs a single task without opening the interface
jcode --resume <name>	Resumes a previous session by its memorable name
jcode serve	Runs jcode as a persistent background server
jcode connect	Connects an additional client to a running server
jcode dictate	Sends voice input from your configured speech to text command
jcode login --provider <name>	Logs in to an AI provider such as Claude, OpenAI, or Gemini
jcode browser status	Checks whether browser automation is set up
jcode browser setup	Sets up browser automation
/model	Switches the active model during a session
/account	Switches between multiple accounts on the same provider
/alignment	Switches between left aligned and centered display mode

These cover most day to day use. The full cheat sheet page expands on flags, shortcuts, and less common commands for people who want to move faster inside jcode.

Connect a Model Provider to jcode

Jcode does not include its own AI model. It connects to a provider you already use, so you can keep using a subscription you already pay for.

Log in to a provider with one simple command:

jcode login --provider claude
jcode login --provider openai
jcode login --provider gemini
jcode login --provider copilot
jcode login --provider azure

Jcode supports many more providers beyond these four, including OpenRouter, DeepSeek, Fireworks, MiniMax, and several other API compatible services. If you run into rate limits on one account, /account lets you switch to a second account on the same provider without logging out first.

For remote or headless setups such as an SSH session, most OAuth style providers support a version of login that avoids opening a local browser:

jcode login --provider openai --no-browser

The full provider connection guide covers every supported provider, plus the two step scriptable login flow useful for automated setups.

Run jcode with Local Models

If you would rather not depend on a paid subscription, or you want your code to stay entirely on your own machine, jcode can connect to models running locally through Ollama or LM Studio.

Using Ollama:

ollama pull llama3.2
jcode login --provider ollama
jcode --provider ollama --model llama3.2 run "hello"

Using LM Studio, after starting its local server and loading a chat model:

jcode login --provider lmstudio
jcode --provider lmstudio --model "<model-id>" run "hello"

Both tools expose the same style of API, so jcode talks to them the same way it talks to remote providers, including streaming responses and tool calling.

Jcode can also connect to almost any custom endpoint that speaks the same standard API, including self hosted setups like vLLM, using the jcode provider add command.

The full local models guide walks through this setup along with config file options for people who prefer editing files directly over using the login screen.

Add MCP Servers to jcode

MCP servers give jcode extra tools beyond its built in ones, such as connecting to a file system, a database, or another application.

Jcode reads MCP configuration from its own config files:

~/.jcode/mcp.json

for servers available everywhere, and:

.jcode/mcp.json

for servers scoped to a single project.

A basic entry looks like this:

{
  "mcpServers": {
    "filesystem": {
      "command": "/path/to/mcp-server",
      "args": ["--root", "/workspace"],
      "env": {},
      "shared": true
    }
  }
}

If you already use Claude Code, jcode automatically tries to import your existing MCP servers from its config files the first time it runs, so you do not need to set them up twice.

Currently jcode supports command based MCP servers directly. Servers that use HTTP or SSE connections are recognized but skipped, with a note logged so you know they were not loaded.

The full MCP guide covers project level configuration, import behavior from other tools, and common setup issues.

Browser Automation in jcode

Jcode includes a browser control tool that lets an agent open pages, read their content, click, type, and take screenshots, all from inside a normal session.

Set it up with:

jcode browser status
jcode browser setup

The current backend is Firefox, connected through Firefox Agent Bridge. Once setup is complete, you can simply ask your agent to open a page, fill out a form, or capture a screenshot, and it will use the tool directly without any extra commands from you.

The interface also summarizes what the browser tool is doing in a short readable line, such as opening a URL or clicking a button, without echoing anything sensitive you type into a form field.

The full browser automation guide covers setup in more detail, what each built in action does, and what to check if the connection is not working.