Memory for coding agents
Instruction files tell a coding agent how to write code. How to give coding tools a memory of decisions and reasons that lasts across sessions and tools.
For developers
A hosted MCP memory server you can actually read
Memory Bridge exposes your memory to any MCP client over a single remote endpoint. No local process, no vector database to run, and the notes it stores are plain text you can open.
The short answer
Memory Bridge is a remote MCP server (Streamable HTTP) with OAuth sign-in. Add its URL to any client that supports remote MCP servers, authorise it, choose which projects it can reach, and the client gets tools to search, read, write and organise notes. ChatGPT, Claude and Codex are connected this way today.
What the server offers
| Area | Detail |
|---|---|
| Transport | Remote MCP over Streamable HTTP at a single endpoint. No stdio process to install. |
| Sign-in | OAuth. The client opens a browser window, you sign in, and the client receives a short-lived token. Tokens expire and are refreshed by the client. |
| Scope | Per connection, you choose the projects it can reach and whether it can read only or read and write. You can revoke a connection at any time. |
| Tools | Search by meaning across your notes, read a note, write or edit a note, list and organise projects, and read earlier versions. Results come back once, without repeated fields, to keep context small. |
| Storage | Notes are plain text with a source line (who, when, which client). Every edit keeps a version. Project notes can be downloaded as files. |
| Language | Search handles Arabic and English, including Arabic spelling variants, so a note written in one wording is found by another. |
| Hosting | Hosted service. Memory files and their history are stored in Riyadh, Saudi Arabia. Nothing to self-host. |
Which clients work
Any MCP client that supports remote servers with OAuth. ChatGPT (apps and connectors), Claude (connectors) and Codex (through the ChatGPT account) have recorded test runs. Claude Code, Cursor, Gemini CLI, Windsurf, Cline, Kiro, OpenCode, Goose, LM Studio and others add remote servers from their settings or config files.
Each client has a page under supported tools with the steps and an example of what the memory adds for that tool.
Connect in three steps
- 1
Add the server
In your client's MCP or connector settings, add the Memory Bridge server URL from your account page.
- 2
Authorise
Sign in when the browser opens, then pick the projects this client may reach and its permission level.
- 3
Use the tools
Ask the client to save a decision or search your memory. It calls the server's tools; you see every write in the activity log.
How it differs from CLAUDE.md, rules files and memory APIs
Instruction files in a repository are read in full at the start of every session and belong to one project on one machine. Memory Bridge is searched on demand, returns only what the request needs, and is shared across projects, machines and clients.
Memory APIs such as Mem0 are infrastructure for memory inside products you build for your users. Memory Bridge is memory for you, the person using the tools. There is no SDK; the MCP tools are the interface. See the comparison with Mem0.
Good to know
- The trial is invite-only. The server URL and the account page come with your invitation.
- Rate and storage quotas apply per account. The server tells the client what to do next when a limit is reached, instead of failing silently.
- Concurrent edits are detected: a client writing over a note that changed since it last read it is asked to re-read first.
- Memory Bridge does not train models on your notes, and semantic indexing runs on its own server.
- Built for personal use today; team workspaces and roles are not available.
Common questions
Is there a REST API or SDK?
No. The MCP tools are the interface, and they are what the connected clients use. If you need a programmatic memory layer inside your own product, a memory API such as Mem0 is the better fit.
Can I self-host it?
Not at present. Memory Bridge is a hosted service with memory stored in Riyadh.
Does it work with local models?
Yes, through any MCP-capable local client, such as LM Studio. The model runs on your machine; the memory is remote.
What does the client see?
Only the projects you allowed, with the permission you chose. Tool results carry the note text and its source line, nothing about other projects.
Keep reading
Works with the AI tools you already use
Memory Bridge works with the AI tools you already use: ChatGPT, Claude, Gemini, Cursor, Claude Code, Codex and more. One AI memory across all of them.
Your memory is yours
Plain answers: what we store, where, who can reach it, what goes out to AI tools, and how to correct your data or ask for it to be destroyed.
One memory. Every tool you use.
The trial is invite-only for now. Request access and we will write to you when your seat opens.