infernoflow — Claude Code MCP server: what it does, how to install
Keeps track of gotchas, decisions, and bookmarks across AI coding sessions using local JSONL files inside your repository.
claude mcp add --env INFERNOFLOW_PROJECT_DIR=<INFERNOFLOW_PROJECT_DIR> --transport stdio infernoflow -- npx -y infernoflowWhat infernoflow does
Fresh AI coding sessions start cold, forcing the agent to rediscover the same project quirks repeatedly. infernoflow solves this by keeping a structured record in .ai-memory/ on disk. No cloud services are required. Branch-specific notes stay in git, so team members inherit the context during pull requests.
The server monitors error markers in prompts and logs them via the amp_write MCP tool. For major milestones, infernoflow bookmark condenses the prior 40 dialogue turns into markdown without spending model tokens. Running infernoflow init also installs a dedicated memory-keeper subagent and skill into .claude/.
Any memory linked to modified code gets flagged as potentially stale until cleared with infernoflow resolve <id>. Setting the INFERNOFLOW_PROJECT_DIR environment variable is required when configuring clients like Cursor or Codex.
What it helps with
- retain architectural decisions across chat restarts
- share discovered API quirks with teammates via git
- drop session bookmarks before starting risky code changes
- retire outdated context when referenced files change
Example prompts for the agent
- Drop a bookmark before we refactor the billing service
- Check where we left off using infernoflow resume
- Retire the outdated note using infernoflow resolve <id>
How to install infernoflow
Platform CLI
claude mcp add --env INFERNOFLOW_PROJECT_DIR=<INFERNOFLOW_PROJECT_DIR> --transport stdio infernoflow -- npx -y infernoflowReplace <…> with your own values
Platform CLI
codex mcp add infernoflow --env INFERNOFLOW_PROJECT_DIR=<INFERNOFLOW_PROJECT_DIR> -- npx -y infernoflowReplace <…> with your own values
Config file
{
"mcpServers": {
"infernoflow": {
"command": "npx",
"args": [
"-y",
"infernoflow"
],
"env": {
"INFERNOFLOW_PROJECT_DIR": "<INFERNOFLOW_PROJECT_DIR>"
}
}
}
}Replace <…> with your own values
Config file
{
"servers": {
"infernoflow": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"infernoflow"
],
"env": {
"INFERNOFLOW_PROJECT_DIR": "<INFERNOFLOW_PROJECT_DIR>"
}
}
}
}Replace <…> with your own values
Platform CLI
gemini mcp add -e INFERNOFLOW_PROJECT_DIR=<INFERNOFLOW_PROJECT_DIR> infernoflow npx -y infernoflowReplace <…> with your own values
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 infernoflow
How to install infernoflow in Claude Code?
Run claude mcp add infernoflow -- npx -y infernoflow in your terminal to set up the MCP server.
Where is the session memory stored?
Entries are written to .ai-memory/branches/<branch>.jsonl for branch-specific notes and .ai-memory/global.jsonl for personal preferences. Everything stays in local files.
How do bookmarks work in infernoflow?
They capture a named checkpoint by summarizing the last 40 conversation turns into markdown deterministically. You can trigger them from the CLI, chat hooks, or the amp_bookmark tool.
