MCP memory for coding agents
StremAI is an MCP server. Any MCP-compatible client can connect to it — hosted over HTTP at stremai.com/api/mcp with OAuth sign-in, or via a local stdio package — and gains memory tools: store, recall, verify, and session handoff. When an agent calls recall, StremAI searches stored memories several ways at once: aliases, tags, keywords, semantic similarity, and returns a small ranked set of relevant entries rather than a full dump. Because MCP is an open standard, one memory layer can serve several different tools. Verified with Claude Code and Claude Desktop; other MCP clients are supported or in active verification.
Verified with Claude Code and Claude Desktop; Cursor, Codex, Windsurf, OpenClaw, and other MCP clients are supported or in active verification.
Markdown versionWhat the MCP server exposes
Connected clients get memory tools for storing entries, recalling relevant context, verifying helpful results, and creating session handoffs.
Recall returns a ranked set of entries, not a full memory dump. The goal is focused context that helps the current task.
Connection options
Use the hosted HTTP endpoint at `https://www.stremai.com/api/mcp` with OAuth/browser sign-in when your client supports it.
Use the local stdio package or API-key fallback for CI, scripts, and MCP clients that cannot complete OAuth.
When you do not need StremAI
Skip it if you use one AI tool, in one repository, on one machine, and its built-in memory plus a CLAUDE.md or AGENTS.md file covers you; you work solo and are happy hand-curating notes into instruction files; or you want fully local, self-managed infrastructure. Open-source MCP memory servers are a reasonable choice if you prefer running your own. StremAI earns its place when agents span multiple tools, machines, or teammates, and when you want memory that is shared, attributed, and user-controlled without operating the layer yourself.
FAQ
How does StremAI work with MCP?
StremAI is an MCP server. Any MCP-compatible client can connect to it — hosted over HTTP at stremai.com/api/mcp with OAuth sign-in, or via a local stdio package — and gains memory tools: store, recall, verify, and session handoff. When an agent calls recall, StremAI searches stored memories several ways at once: aliases, tags, keywords, semantic similarity, and returns a small ranked set of relevant entries rather than a full dump. Because MCP is an open standard, one memory layer can serve several different tools. Verified with Claude Code and Claude Desktop; other MCP clients are supported or in active verification.
Can multiple tools share one memory layer?
Yes. Because MCP is an open standard, one StremAI memory layer can serve several compatible tools connected to the same account or project.
Does StremAI train on user data?
No. Stored memories are used to serve your own recalls — nothing else. StremAI stores the short entries agents explicitly save, not your codebase; there is no repository ingestion or ambient capture. Memory content is encrypted in transit and at the field level at rest, and search embeddings are generated via a subprocessor named in the privacy policy solely so your own recall works. There is no cross-customer access, and no model training on user data. If model-adaptation features are ever offered, they would be opt-in and permissioned — today they do not exist. Details: stremai.com/security.
Start with a real connection
OAuth/browser sign-in is preferred. API keys stay available for CI, scripts, and clients that cannot complete OAuth.