Supports signing in to the hosted ASR Memory server with a GitHub account via OAuth 2.1, letting users connect their AI tools without copying an API key.
Supports signing in to the hosted ASR Memory server with a Google account via OAuth 2.1, letting users connect their AI tools without copying an API key.
Supports signing in to the hosted ASR Memory server with a Kakao account via OAuth 2.1, letting users connect their AI tools without copying an API key.
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., "@ASR Memory Save to memory: we chose Postgres for billing."
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.
ASR Memory is shared memory for your AI tools, run as a hosted remote MCP server. Save a decision in one tool and find it, word for word, from another.
This repository holds documentation, agent instructions and examples. The server source is not public.
Website: https://asrmemory.com
MCP endpoint: https://asrmemory.com/mcp (Streamable HTTP, OAuth 2.1)
Official MCP Registry: com.asrmemory/asr
Add https://asrmemory.com/mcp as a remote MCP server (custom connector) in your AI tool.
Sign in when the login page opens (email, Google, Kakao or GitHub). No key to copy.
Ask the tool: "Save to ASR: we chose Postgres for billing." Then, in another tool: "What did we decide about billing?"
Tools that cannot open a login page can connect with an API key issued in the dashboard (https://asrmemory.com/app/). Step-by-step guides for each client: https://asrmemory.com/en/docs.html
To see the project dashboard without signing up, open the demo: https://asrmemory.com/madang/?demo
Related MCP server: Sovereign Universal Memory MCP
Tool | What it does |
| Save a memory as written, with who saved it and when it happened |
| Search by meaning and keywords |
| Show one memory in full |
| Recall what happened on a date |
| Brief a new session on recent work |
| Counts by drawer and tool |
| Change a memory's drawer |
| Move a memory to the trash |
| List the trash |
| Restore from the trash |
| Permanently delete (asks first) |
| Export your memories as JSON Lines |
| Store and read an encrypted value |
| Track a goal day by day; only you mark the result |
| Send a task to another tool |
| Hand off unfinished work when a session ends |
| See and take tasks sent to you |
| Close a task with its result |
| Save a project status summary |
AGENTS.md is a set of rules to paste into CLAUDE.md, AGENTS.md or GEMINI.md so long-running agents read memory at the start, save decisions as they go, and hand off work between tools. It also works without ASR. Korean: AGENTS.ko.md, Japanese: AGENTS.ja.md.
For chat apps (claude.ai project instructions, ChatGPT custom instructions), use the short 11-line version: https://asrmemory.com/en/docs.html#guide
Each account has its own memory space. Cross-account access tests run on every deployment, and the deployment stops if any fails.
Deleted memories go to the trash first and can be restored. Permanent deletion is a separate, explicit step.
You can export everything with memory_export and delete your account from the dashboard.
Where data goes (database, server, embedding and classification providers): https://asrmemory.com/en/privacy.html
Do not store passwords, keys or children's conversations as plain memories. Use secret_save for sensitive values.
Free during the beta, up to 1,000 memories per account. Searching, recall and export are not counted. Project summaries saved with asr_tidy do not count toward the limit.
Open an issue in this repository, or email idoweddings@naver.com.
Documentation and agent instructions in this repository: CC BY 4.0. Example code in examples/: MIT. See LICENSE.
This server cannot be deployed
AI memory layer — one shared, persistent memory across every AI tool you connect.
Governed memory and workspace for any AI: tasks, calendar, mail and pages, with per-action consent.
Cross-model memory for people and teams: one memory every AI you use loads at session start.
Persistent memory for AI agents. Semantic search, memory graph, W3C DID identity.
