Local semantic code search MCP server that gives Claude Code, Codex, Cursor and OpenCode meaning-based lookup over your codebase using Ollama or LM Studio embeddings.
ykdojo/safeclaw
Runs multiple Claude Code sessions, each isolated in its own Docker container, managed from a web dashboard.
- Stars
- 186
- Forks
- 20
- Open issues
- 0
- Last commit
- last month
What is it?
SafeClaw packages Claude Code so each session runs in its own container with bypass permissions enabled, keeping the host machine out of reach. Sessions can be created, started, stopped, and deleted in seconds, and each one is reachable through a web terminal built on ttyd and tmux.
A Node.js dashboard lists all sessions with start, stop, and delete controls, lets you set volume mounts and initial queries when creating one, and shows live iframe views of active sessions. Conversation history and auto memory are stored on the host under ~/.config/safeclaw/sessions/, so rebuilding containers does not remove them.
Setup uses shell scripts for building the image and starting named sessions, with tokens such as CLAUDE_CODE_OAUTH_TOKEN and GH_TOKEN injected from a local secrets directory. The image includes GitHub CLI, Playwright MCP with Chromium, and a set of skills for Slack reading, yt-dlp downloads, and reading public Google Docs.
Key features
- 01Each Claude Code session runs in an isolated Docker container
- 02Web dashboard to create, start, stop, and delete sessions
- 03Web terminal per session via ttyd and tmux
- 04Conversation history and memory persist on the host through volume mounts
- 05Named sessions and volume mounts configured from the command line
Use cases
- Running several Claude Code tasks in parallel without them interfering with each other
- Letting Claude Code use bypass permissions on a project while keeping the host protected
- Keeping per-project research or experiments in separate, disposable environments
Quick start
- 1
Build the Docker image
./scripts/build.sh - 2
Start a container and the web terminal
./scripts/run.sh - 3
Start the web dashboard to manage sessions
node dashboard/server.js
Things to know
- Requires Docker to be installed and running on your machine.
- Needs a Claude Code token generated with claude setup-token before the first run.
In collections
Topics
FAQ
- What is safeclaw?
- Runs multiple Claude Code sessions, each isolated in its own Docker container, managed from a web dashboard. SafeClaw packages Claude Code so each session runs in its own container with bypass permissions enabled, keeping the host machine out of reach. Sessions can be created, started, stopped, and deleted in seconds, and each one is reachable through a web terminal built on ttyd and tmux.
- Is safeclaw free to use? What license does it have?
- safeclaw is open source under the MIT license. Check the license terms before using it in commercial products.
- Is safeclaw still maintained?
- Updated within 6 months. The last commit to safeclaw was on Sep 27, 2026.
- What can I use safeclaw for?
- Running several Claude Code tasks in parallel without them interfering with each other Letting Claude Code use bypass permissions on a project while keeping the host protected Keeping per-project research or experiments in separate, disposable environments
- How do I install safeclaw?
- Build the Docker image: ./scripts/build.sh Start a container and the web terminal: ./scripts/run.sh Start the web dashboard to manage sessions: node dashboard/server.js
- Where can I get safeclaw?
- The source code and installation instructions are on GitHub: https://github.com/ykdojo/safeclaw
Summary generated by AI from the project's README. GitHub stats refresh daily.
Related projects
A Claude Code skill set that turns a GitHub Projects kanban board into an autonomous build, QA and review loop that returns merge-ready PRs.
A starter configuration and CLAUDE.md memory-bank template for organizing Claude Code settings, hooks, subagents and skills in a project.
A local MCP server that lets AI agents search arXiv, read papers section by section from original LaTeX, export BibTeX, and watch topics.
Multi-agent orchestration layer for Claude Code, adding team pipelines, autonomous execution and tmux-based CLI workers.
A self-hosted, Web UI-based personal AI coworker powered by a Claude Code agent, with IM integration and sandboxed task execution.