Local Coding Agent — Claude Code MCP server: local tasks
Hands narrow coding jobs to a model running on your own machine. A patch lands only after a real test run confirms it.
claude mcp add local-coding-agent -- uvx local-coding-agentWhat Local Coding Agent does
Local Coding Agent is an MCP server and safety layer that lets a cloud coding agent pass a small task down to a model on your hardware. The caller supplies a goal, a workspace, which files may be touched, and which checks to run. The local model drafts a change; the controller decides whether it holds up.
A change counts as applied only if the edit stayed inside the declared files, cleared static analysis, and an external test runner confirmed the configured checks. When a post-apply check fails, the edit is rolled back, and that rollback is compared against recorded file digests. The local model never gets open shell or write access to your workspace.
Running it takes Python 3.10 or newer, Git on PATH, and Ollama or an OpenAI-compatible local model server — live responses need one of those. Commits, tags and releases stay in your hands; the agent does not create them. There is a local-agent CLI and a Windows Desktop Harness.
What it helps with
- delegate a small fix to a local model without exposing the whole repo
- gate a patch behind an external test run before it applies
- roll back a change when post-apply checks fail
- split a large request into smaller coding tasks
- wire a local model into a cloud agent over MCP
Example prompts for the agent
- Connect Local Coding Agent and hand the local model a task: fix error handling in
src/api.py, check withpytest tests/test_api.py - Break my request into smaller tasks with
local-agent decomposeand show the result - Run
local-agent doctorto confirm the local backend is ready
How to install Local Coding Agent
Platform CLI
claude mcp add local-coding-agent -- uvx local-coding-agentuvx runs the command named after the package; if the README names another one, use uvx --from local-coding-agent <command>
Platform CLI
codex mcp add local-coding-agent -- uvx local-coding-agentuvx runs the command named after the package; if the README names another one, use uvx --from local-coding-agent <command>
Config file
{
"mcpServers": {
"local-coding-agent": {
"command": "uvx",
"args": [
"local-coding-agent"
]
}
}
}uvx runs the command named after the package; if the README names another one, use uvx --from local-coding-agent <command>
Config file
{
"servers": {
"local-coding-agent": {
"type": "stdio",
"command": "uvx",
"args": [
"local-coding-agent"
]
}
}
}uvx runs the command named after the package; if the README names another one, use uvx --from local-coding-agent <command>
Platform CLI
gemini mcp add local-coding-agent uvx local-coding-agentuvx runs the command named after the package; if the README names another one, use uvx --from local-coding-agent <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 Local Coding Agent
How do I install Local Coding Agent in Claude Code?
Add the server with claude mcp add local-coding-agent -- uvx local-coding-agent. You need Python 3.10 or newer and Git on PATH first.
What does Local Coding Agent do?
It lets a cloud agent assign a narrow task to a local model and apply the patch only once an external test run has confirmed it. The rollback path is verified against recorded file digests too.
What do I need for live model responses?
Install Ollama or run an OpenAI-compatible local model server. Without one, inference stays unavailable.
