[Purpose] StremAI is a shared memory layer for AI coding agents. It lets connected agents in tools such as Claude Code, Cursor, and Codex save what they learn during work, and other connected agents can later recall this content across sessions, machines, and tools. In short, it provides "one brain" for all coding agents.
[Main Capabilities]
- Share project context across sessions, machines, and tools, avoiding the need to re-explain the same repository, conventions, or decisions every time.
- Store the "learning layer" content accumulated by agents during work, including: decisions and architectural context, environment setup pitfalls and repository-specific commands, traps that cost time and should not be hit again, handoff information between Claude Code/Cursor/Codex and teammates, and project facts that should be provided to the next connected agent.
- Memory is user-controlled: entries are human-readable, labeled with the agent that stored them, and can be exported, archived, or deleted from StremAI.
- Division of labor with user-written instruction files (CLAUDE.md, AGENTS.md, project README): the latter are instructions written by the user, while StremAI stores memories learned by agents during work.
[How to Use]
- Hosted MCP preferred: connect in an MCP client that supports OAuth/browser login https://stremai.com/api/mcp .
- Connect Claude Code: after running the command, approve the login in the browser. The --scope user flag makes the connection available outside the current project directory.
- Connect other MCP clients: use the hosted endpoint above; if the client cannot complete OAuth/browser login, create an API key in StremAI and use a Bearer header.
- API key as fallback: suitable for CI, scripts, or clients that cannot complete OAuth/browser login.
- Local package fallback: suitable for clients that only support stdio or for local experiments. The package name is a legacy name kept for compatibility; new documentation and the server alias use stremai.
- Compatibility names retained: some packages and tools still use the original agentbay or aiagentsbay-mcp names, so that existing installations continue to work.
[Applicable Scenarios]
- Every new agent session requires re-explaining the same repository, conventions, or decisions.
- Context disappears between sessions or after switching machines.
- One agent learns something (a pitfall, a fix, an architectural decision), but other agents know nothing about it.
- Handoff notes are outdated, and there is nothing to tell the agent which version is still applicable.
- You use multiple coding tools and want them to share context instead of starting from scratch.
[Tool Compatibility] Claude Code, Cursor, Codex, Windsurf, OpenClaw, Hermes, and other MCP clients can use the same memory layer.
[Limitations and Notes] The hosted application and some SDK implementation details are developed separately. This repository is StremAI's public installation, comparison, scorecard, and MCP recipe page; the repository name retains agentbay for compatibility with old links and package metadata. The repository content is licensed under MIT.
[Entry Points] Website: stremai.com MCP documentation: stremai.com/docs/mcp-memory Claude Code memory: stremai.com/docs/claude-code-memory Python SDK: stremai.com/docs/python-sdk AI answer manifest: stremai.com/llms-full.txt
