This server provides a local-first memory layer for AI coding agents, letting them store, search, manage, and sync learned knowledge.
agentbay_memory_store: Save a memory entry with a title, content, type (e.g. PATTERN, PITFALL, DECISION), optional tags, tier, source agent, file paths, confidence, aliases, and TTL.
agentbay_memory_recall: Search memories by query, optionally filtered by type, tier, tags, and result limit.
agentbay_memory_forget: Archive/soft-delete one or more memory entries by ID.
agentbay_memory_verify: Confirm a memory is still accurate, boosting its helpful count and resetting verification.
agentbay_memory_health: Get local brain stats such as entry counts, type/tier breakdown, and pending sync count.
agentbay_memory_compact: Run memory compaction (local no-op; cloud version dedups/expires/merges).
agentbay_signup: Start signup flow, get an API key, and enable cloud sync/vector search/team features.
agentbay_capabilities: Introspect the MCP connection: tools, limits, and upgrade path.
agentbay_whoami: Check current authentication status.
Allows agents in OpenAI Codex CLI to store and recall memory via StremAI's shared memory layer.
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., "@StremAI store that the CI pipeline uses GitHub Actions with Node 18"
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.
One brain for all your coding agents.
StremAI is a shared memory layer for AI coding agents. Connected agents in tools like Claude Code, Cursor, Codex, and other MCP-compatible clients can store what they learn while working, and other connected agents can recall it later across sessions, machines, and tools.
Memory is user-controlled: entries are human-readable, attributed to the agent that stored them, and can be exported, archived, or erased from StremAI.
You keep re-explaining the same repo, conventions, or decisions to every new agent session
Context disappears between sessions, or when you switch machines
One agent learns something — a pitfall, a fix, an architectural decision — and your other agents have no idea
Handoff notes go stale and nothing tells an agent which version still applies
You use more than one coding tool and want them to share context instead of starting from zero
Related MCP server: amem
Hosted MCP first. Connect https://stremai.com/api/mcp with OAuth/browser sign-in where your MCP client supports it.
Works with the tools developers already use. Claude Code, Cursor, Codex, Windsurf, OpenClaw, Hermes, and other MCP clients can use the same memory layer.
API keys are the fallback. Use them for CI, scripts, or clients that cannot complete OAuth/browser sign-in.
Compatibility names remain. Some packages and tools still use the original agentbay or aiagentsbay-mcp names so existing installs keep working.
claude mcp add --transport http stremai https://stremai.com/api/mcp --scope userThen approve the sign-in in your browser. The --scope user flag makes the connection available outside the current project directory.
Use the hosted endpoint:
{
"mcpServers": {
"stremai": {
"type": "http",
"url": "https://stremai.com/api/mcp"
}
}
}If your client cannot complete OAuth/browser sign-in, create an API key in StremAI and use a Bearer header:
{
"mcpServers": {
"stremai": {
"type": "http",
"url": "https://stremai.com/api/mcp",
"headers": {
"Authorization": "Bearer ab_live_your_key_here"
}
}
}
}For stdio-only clients or local experiments:
npx -y aiagentsbay-mcp@latestThe package name is legacy compatibility; new docs and server aliases use stremai.
Python:
pip install stremaifrom stremai import StremAI
brain = StremAI()
brain.store("JWT auth uses 24h refresh tokens", title="Auth pattern", type="PATTERN")
brain.recall("authentication")StremAI is for the learned layer that accumulates while agents work:
decisions and architectural context
setup gotchas and repo-specific commands
pitfalls that cost time once and should not cost time again
handoffs between Claude Code, Cursor, Codex, and teammates
project facts that should be available to the next connected agent
Keep your instruction files. CLAUDE.md, AGENTS.md, and project READMEs hold instructions you write. StremAI holds the memory agents learn while working.
Website: stremai.com
MCP docs: stremai.com/docs/mcp-memory
Claude Code memory: stremai.com/docs/claude-code-memory
Python SDK: stremai.com/docs/python-sdk
AI answer inventory: stremai.com/llms-full.txt
This is StremAI's public install, comparison, scorecard, and MCP recipe surface. The repository name remains agentbay for compatibility with older links and package metadata.
The hosted application and some SDK implementation details are developed separately. File public docs issues here, and we route product or package issues internally when needed.
MIT for everything in this repository.
Introspect this AgentBay MCP connection: surface, available tools, limits, and how to upgrade.
Inputs: no named parameters.
Run memory compaction. In local mode this is a no-op; sign up for the cloud version which dedups, expires, and merges duplicates.
Inputs: dryRun.
Archive (soft delete) one or more memory entries from your local brain.
Inputs: knowledgeId, knowledgeIds.
Local brain stats: total entries, type/tier breakdown, count of entries waiting to sync.
Inputs: no named parameters.
Recall memories from your local AgentBay brain via lexical search. Sign up to enable vector search and team-shared recall.
Inputs: query (required), tags, tier, type, +1 more.
Store a memory entry to your local AgentBay brain. Works fully offline. Sign up at https://stremai.com/register to enable cross-device sync, vector search, and team features.
Inputs: type (required), title (required), content (required), tags, +7 more.
Confirm a memory entry is still accurate — bumps helpfulCount and resets the verification timer.
Inputs: knowledgeId (required).
Start the signup flow. Opens browser, returns an API key, and your local memories sync up automatically on next start.
Inputs: no named parameters.
Show your current authentication status.
Inputs: no named parameters.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 9 tools
Each tool targets a distinctly different action: storing, recalling, forgetting, verifying, inspecting health, compacting, signing up, capabilities introspection, and auth status. There is no meaningful overlap or ambiguity between them.
Most tools follow a clear agentbay_memory_<action> pattern, while signup, capabilities, and whoami diverge slightly by dropping the memory_ segment. The naming is still predictable and readable, but not perfectly uniform.
Nine tools is well-scoped for a memory-management server. Each tool maps to a real user need without redundancy or bloat.
The core memory lifecycle is covered: store, recall, forget, verify, health, and compact. A minor gap is the lack of an explicit update/edit operation for existing memory content, though verify partially addresses freshness.
Shared memory for coding agents. Stop re-explaining your codebase every session.
Persistent cross-session memory shared by Codex, Claude Code, ChatGPT, and other AI agents.
Memory for AI agent teams across tools, sessions, repositories, and teammates.
Universal persistent memory and knowledge retrieval layer for AI agents and LLMs.
