bayma
Provides durable REPL sessions using the Bun runtime, allowing stateful execution of JavaScript/TypeScript code with persistent state and recovery from checkpoints.
Provides durable REPL sessions using Python, enabling stateful execution of Python code with persistent state and recovery from checkpoints.
Provides durable REPL sessions using Rust, enabling stateful execution of Rust code with persistent state and recovery from checkpoints.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@baymaStart a Python session and compute the first 10 Fibonacci numbers."
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
bayma
A durable engine for REPL sessions. bayma allows agents to run and manage long-lived, Jupyter-like code execution sessions. For an agent, an individual bayma REPL session works like a Jupyter notebook, presented to it as an MCP server whose function calls look similar to other "code mode" execution interfaces.
Get Started
On Linux (x64), with Docker, pull bayma's image, which holds everything bundled with bayma: its runtimes, their toolbelt, and what it takes to keep REPL sessions on disk.
docker pull ghcr.io/eaucoin/baymaThen add bayma to your agent setup as an MCP server, for Claude Code or for Codex:
BAYMA='exec docker run -i --rm --user "$(id -u):$(id -g)" -e HOME -v "$HOME:$HOME" -w "$PWD" --cap-add CHECKPOINT_RESTORE --cap-add SYS_PTRACE --security-opt seccomp=unconfined ghcr.io/eaucoin/bayma'
claude mcp add bayma -- sh -c "$BAYMA"
codex mcp add bayma -- sh -c "$BAYMA"For any other MCP client, use {"command": "sh", "args": ["-c", "<$BAYMA>"]}.
bayma runs as you, with your home directory at its own path, and the last
three flags let it snapshot REPL sessions within its own container.
docker run --rm ghcr.io/eaucoin/bayma doctor checks that each runtime works.
REPL sessions are kept under ~/.local/state/bayma, one directory per project
directory the client launched from. When bayma stops, it snapshots each idle
REPL session, its whole process and everything it holds, and a later bayma
restores it where it left off; after a reboot, a REPL session comes back from
its own checkpoints instead.
Then you can use bayma. Each runtime has a skill, bayma-runtime-*, on how to
work in that runtime with bayma. Within a REPL session, an agent executes code
in that runtime, including against legacy software. For example, in the C++
runtime, with libardour, it can run a headless digital audio workstation:
npx skills add eaucoin/bayma --skill bayma-runtime-cpp bayma-platform-ardouror with MLT++, a headless non-linear editor for video:
npx skills add eaucoin/bayma --skill bayma-runtime-cpp bayma-platform-mltppOr in the Python runtime, with bpy, it can model, animate, and render 3D scenes in a headless Blender:
npx skills add eaucoin/bayma --skill bayma-runtime-python bayma-platform-bpyEach of these is a bayma-platform-* skill: one per platform, meaning an
application or service an agent drives through its API. The skill shows where
that API's reference is, and keeps what the platform needs in its own folder
rather than being bundled with bayma.
Within each of the two namespaces, the skills share a similar structure, which makes it easier to add new runtimes and platforms.
Related MCP server: mcp-python-repl
Development
See package.json for the development scripts, Dockerfile for the image,
.github/workflows/ for the GitHub Actions workflows, and
.agents/skills/development-observability/ for development observability.
This server cannot be deployed
Maintenance
Related MCP Connectors
Hosted runtime for persistent agent teams, durable workflows, memory, schedules, and goals.
Hosted MCP memory and agent control plane for durable conversations, jobs, and operations.
System-of-record notebook for AI coding agents: pages, datastores, tasks, skills over MCP.
Hosted MCP memory for coding agents: persistent across sessions, editable markdown, team sharing.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceThe sessionless code interpreter. Securely run AI-generated code in stateful sandboxes that run forever.93 npm229MIT
- AlicenseAqualityDmaintenanceA production-grade MCP server providing a persistent Python REPL with multi-session support, sandboxing, and timeout protection, enabling LLM agents to execute Python code across multiple turns with variables that persist between calls.1221 PyPI1MIT
- AlicenseNot gradedqualityDmaintenanceMCP server for secure, session-based Python code execution in Docker containers, enabling LLM applications to run code, manage state, and access files.9MIT
- FlicenseNot gradedqualityCmaintenanceA runtime for inspectable agent workflows that provides MCP tools, bounded Python execution, session memory, and deterministic evaluation.-