MCP Bifrost runs on your own machine — the client starts it, so there is no endpoint to ping. 115 installs a week from pypi. Last commit 29 Aug 2026.
Delegate bulk code edits to a cheap worker model. Gated, atomic. PHP and Python.
Every tool that appeared, vanished or quietly changed what it asks for. Recorded since 14 September 2026. No other catalogue keeps this.
Today is the operative word: we check MCP Bifrost every 15 minutes and re-read its code on every release. Watch it and you find out the day that stops being true.
This server runs on your own machine — install it with the package manager and the client starts it for you. Package name taken from the official registry entry.
claude mcp add mcp-bifrost -- uvx mcp-bifrost
{
"mcpServers": {
"mcp-bifrost": {
"args": [
"mcp-bifrost"
],
"command": "uvx"
}
}
}
[mcp_servers.mcp-bifrost]
command = "uvx"
args = ["mcp-bifrost"]
{
"mcpServers": {
"mcp-bifrost": {
"args": [
"mcp-bifrost"
],
"command": "uvx"
}
}
}
{
"mcpServers": {
"mcp-bifrost": {
"args": [
"mcp-bifrost"
],
"command": "uvx"
}
}
}
This one needs environment variables set before it will start:
DEEPSEEK_API_KEY (Worker API key. Not needed when BIFROST_WORKER_BASE_URL points at a local endpoint.), BIFROST_WORKER_BASE_URL (OpenAI-compatible endpoint for the worker model. Set this to run a local model so no code leaves the machine.), BIFROST_WORKER_MODEL (Worker model identifier. Defaults to deepseek-chat.), BIFROST_DB (Path to the operational log. Defaults to .bifrost/history.db in the working directory.).
The author declared them in the registry entry; get the values from the project itself.
Delegate real-world actions from Claude Code, Codex, and other agents to Hermes Agent.
Local LLM does Claude's bulk work: reads logs and files, researches the web, writes test-gated code.
Auto-detect stubbable packages for Python exe builds and generate minimal stub code
Token-aware ripgrep code search and atomic multi-file edits for coding agents.
Cell fabric daemon: spawn and coordinate Claude/Codex/pi worker cells over MCP. Audited.
Coded error stack trace debugger and automated patch generator.
Discover funded agent work and prepare approval-gated TaskMarket delegations.
Generates unit tests for Python code with coverage before/after reports and concrete edge cases.
Answers built from our own checks of this server.