ClaudeSmalltalk — Claude Code MCP server: what it does, how to install
An MCP server that hands your coding agent a live Squeak or Cuis image to browse, edit and save.
claude mcp add claudesmalltalk -- uvx claude-smalltalkWhat ClaudeSmalltalk does
The server speaks TCP to a running Smalltalk image and gives your agent 14 tools: run an expression, inspect a class, read or replace a method, follow superclasses, save the image. Your agent only calls smalltalk_task. That tool forwards the work to a second model — Ollama on your own hardware, or Anthropic, OpenAI, xAI over the network.
Point the server at a config by setting SMALLTALK_MCP_CONFIG; a starter smalltalk-mcp.json is in examples/. TCP is the default transport, and the VM comes up on first use with a UUID token on each request. Choose mqtt when the image sits behind a broker, or when Cuis runs its MQTT handler.
You need Python 3.10+ and uv, which pulls httpx by itself. A plain python3 setup means installing httpx yourself, plus paho-mqtt if you go with MQTT. On macOS put the VM and the image under /Applications/ or in your home folder: the privacy layer blocks ~/Documents/, ~/Desktop/ and ~/Downloads/, and those paths fail.
What it helps with
- read a method's source without opening a class browser
- define a method from the chat window
- trace where behaviour is declared up the superclass chain
- save the image as a new version number
- list classes by prefix before a refactor
Example prompts for the agent
- Use
smalltalk_list_classesto list every Smalltalk class whose name starts withString. - Show me the superclass chain and immediate subclasses of
String, then save the image as a new version. - Review the code in the image: read the
Stringmethods withsmalltalk_method_sourceand suggest fixes.
How to install ClaudeSmalltalk
Platform CLI
claude mcp add claudesmalltalk -- uvx claude-smalltalkuvx runs the command named after the package; if the README names another one, use uvx --from claude-smalltalk <command>
Platform CLI
codex mcp add claudesmalltalk -- uvx claude-smalltalkuvx runs the command named after the package; if the README names another one, use uvx --from claude-smalltalk <command>
Config file
{
"mcpServers": {
"claudesmalltalk": {
"command": "uvx",
"args": [
"claude-smalltalk"
]
}
}
}uvx runs the command named after the package; if the README names another one, use uvx --from claude-smalltalk <command>
Config file
{
"servers": {
"claudesmalltalk": {
"type": "stdio",
"command": "uvx",
"args": [
"claude-smalltalk"
]
}
}
}uvx runs the command named after the package; if the README names another one, use uvx --from claude-smalltalk <command>
Platform CLI
gemini mcp add claudesmalltalk uvx claude-smalltalkuvx runs the command named after the package; if the README names another one, use uvx --from claude-smalltalk <command>
Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about ClaudeSmalltalk
How do I install ClaudeSmalltalk in Claude Code?
One command: claude mcp add claudesmalltalk -- uvx claude-smalltalk. That registers the server, and the Smalltalk image starts on the first call.
What does ClaudeSmalltalk do?
It connects your agent to a running Squeak or Cuis image so it can browse classes, evaluate code, change methods and save the image. The Smalltalk reasoning runs on a separate model you pick in the config.
Do I need a paid API key?
Not with Ollama, which runs locally at no cost and keeps image code on your machine. The Anthropic, OpenAI and xAI providers are paid, and the key goes into smalltalk-mcp.json.
