# Agentage Memory > One memory. Every AI. Owned by you. This file contains the full public content of https://agentage.io in one document: docs, then blog posts. Index version: https://agentage.io/llms.txt --- # Agentage Memory - Docs Connect any MCP client to one shared markdown memory through a single endpoint. Sign in with OAuth and Claude, ChatGPT, Cursor, and your other AI tools read and write the same notes - plain markdown you own and can export anytime. ## Connect - MCP endpoint: `https://memory.agentage.io/mcp` (Streamable HTTP) - Auth: OAuth 2.1 with PKCE and dynamic client registration - sign in once in the browser, no API key. ### Claude Code 1. Run `claude mcp add --transport http agentage-memory https://memory.agentage.io/mcp` 2. Run `/mcp` and complete the OAuth sign-in. 3. A connector you add in Claude.ai also syncs here automatically. ### Claude (claude.ai & Desktop) 1. One-click: [Add to Claude](https://claude.ai/customize/connectors?modal=add-custom-connector&connectorName=Agentage%20Memory&connectorUrl=https%3A%2F%2Fmemory.agentage.io%2Fmcp) - opens with the server prefilled; review, confirm, and sign in. 2. Or manually: Settings > Connectors > Add custom connector, paste `https://memory.agentage.io/mcp`. ### VS Code 1. One-click: [Install in VS Code](vscode:mcp/install?%7B%22name%22%3A%22agentage-memory%22%2C%22type%22%3A%22http%22%2C%22url%22%3A%22https%3A%2F%2Fmemory.agentage.io%2Fmcp%22%7D). 2. Or add the server to `.vscode/mcp.json` (or run the `MCP: Add Server` command): 3. ```json { "servers": { "agentage-memory": { "type": "http", "url": "https://memory.agentage.io/mcp" } } } ``` 4. VS Code opens the browser for the OAuth sign-in on first use. ### Cursor 1. One-click: [Add to Cursor](cursor://anysphere.cursor-deeplink/mcp/install?name=agentage-memory&config=eyJ1cmwiOiJodHRwczovL21lbW9yeS5hZ2VudGFnZS5pby9tY3AifQ==), then sign in when prompted. 2. Or add the server to `~/.cursor/mcp.json`: 3. ```json { "mcpServers": { "agentage-memory": { "url": "https://memory.agentage.io/mcp" } } } ``` ### ChatGPT 1. Open [Settings > Apps & Connectors](https://chatgpt.com/#settings/Connectors) > Advanced settings > enable Developer Mode > Create (paid plans, not Free; ChatGPT has no one-click link, so add it by hand). 2. Paste `https://memory.agentage.io/mcp` and sign in when prompted. ## MCP tools Once connected, every client gets six memory operations: - `memory__search`: Find notes by keyword across the whole memory. - `memory__read`: Read a note by path. - `memory__write`: Create or replace a note. - `memory__edit`: Apply a targeted edit to a note. - `memory__list`: Browse the folder tree - subfolders with file counts, two levels deep. - `memory__delete`: Remove a note. ## Example prompts - "Search my memory for "postgres" and read the top result." - Outcome: Ranks notes mentioning postgres by match count, then returns the top note in full (frontmatter + markdown body). (`memory__search -> memory__read`) - "Save a note at projects/acme/stack.md: we use Postgres for full-text search." - Outcome: Creates (or fully replaces) that note. It persists server-side and is readable from every AI you connect to the same account. (`memory__write`) - "List everything under projects/ and tag the roadmap note as launched." - Outcome: Shows the folder tree under projects/ (subfolders with file counts + notes), then shallow-merges a launched tag into the roadmap note frontmatter without rewriting the body. (`memory__list -> memory__edit`) The memory tools run only when your request is about your notes - everyday questions like facts, math, or writing are answered normally, without touching your memory. ## REST API An HTTP surface at `https://api.agentage.io`, reads and writes over `/v1/memories`. Auth: OAuth 2.1 bearer token - the same token issued when you connect any MCP client. No API key. memory:read gates reads, memory:write gates writes. Rate limited to 60 requests per minute per IP. - `GET /v1/memories`: List the memories your token can see. - `POST /v1/memories`: Create a memory. Idempotent on an existing name. - `GET /v1/memories/{memory}`: Memory stats: counts, size, rev, last activity. - `DELETE /v1/memories/{memory}`: Retire a memory (recoverable). - `GET /v1/memories/{memory}/notes`: List notes and folders, paginated, filterable by folder/type/tags. - `GET /v1/memories/{memory}/notes/{path}`: Read one note: frontmatter + markdown body. - `PUT /v1/memories/{memory}/notes/{path}`: Create or replace a note. - `PATCH /v1/memories/{memory}/notes/{path}`: Edit a note: replace, append, or str_replace. - `DELETE /v1/memories/{memory}/notes/{path}`: Delete a note (recoverable). - `GET /v1/memories/{memory}/search`: Keyword search, ranked, snippets only. - `GET /v1/memories/{memory}/export`: Stream one memory as a cloneable git bundle. - `GET /v1/memories/export`: Stream every visible memory as one zip. `/v1/vaults/*` still resolves, as a deprecated alias of `/v1/memories/*`, sunsetting Sun, 08 Feb 2027 00:00:00 GMT. It is a path alias, not a shape-compat layer: responses already use the new memories field names. Every response on the alias carries `Deprecation: true`, `Sunset: Sun, 08 Feb 2027 00:00:00 GMT`, and `Link: ; rel="successor-version"`. Full reference: https://docs.agentage.io/rest-api ## Limitations - Notes are plain markdown addressed by path (e.g. work/tasks/foo.md), not by title or ID. - Search is literal keyword/substring matching - not semantic or vector search; search one distinctive keyword, not a phrase. - search returns ranked paths + snippets, never full bodies - use read for the whole note. - list shows the folder tree two levels deep, up to 100 entries per folder - call it again with a subfolder to browse deeper. - edit shallow-merges top-level frontmatter keys and cannot remove a key; use write to fully replace a note. - delete is a recoverable soft-delete (kept in history), not a hard wipe. - One memory per account; connect the same account in each client to share it. Data is stored in the EU and exportable anytime. HTML version: https://docs.agentage.io --- # Your Notes Are Already a Graph. I Turned On the Lights. Published June 23, 2026. Source: https://agentage.io/blog/your-notes-are-already-a-graph Your Obsidian vault is already a graph. [Install Agentage Galaxy](obsidian://show-plugin?id=agentage-galaxy) and one click turns it into a rotating 3D galaxy like the one above. The rest of this piece is why that picture matters. On June 12, 2026, Google Cloud published a spec called the [Open Knowledge Format](https://cloud.google.com/blog/products/data-analytics/how-the-open-knowledge-format-can-improve-data-sharing). It defines knowledge as "a directory of markdown files with YAML frontmatter." Its reference implementation ships a static HTML file that "turns any OKF bundle into an interactive graph view." I have been keeping my notes in exactly that format for years, in Obsidian. A folder of Markdown files, joined by `[[links]]`. So this piece is not a prediction. It is a field report from inside the format Google just standardized. The thesis is simple: your knowledge is already a graph. Plain Markdown plus `[[links]]` is the format both you and every LLM now read and write. A 3D galaxy is just what that graph looks like when you turn on the lights. And this is not a preference anyone can switch off. These models were trained on oceans of Markdown. They take it in and put it back out by default. That is learned behavior baked into the weights, not a vendor toggle or a customer setting. You can ask a model for less of it, but you are fighting the default, not changing it. ## Why is Markdown the format models read and write? Because the labs put that choice in writing. The [llms.txt](https://llmstxt.org/) spec picked Markdown over XML, and it says why: "because we expect many of these files to be read by language models and agents." That is a named standard choosing a format on the explicit grounds that LLMs are the readers. Jeremy Howard proposed it on September 3, 2024. This is not one person's taste. The agent-config layer standardized on the same primitive, at scale. AGENTS.md is a Markdown-based convention adopted by more than 60,000 open-source projects, and it now sits under the Linux Foundation's [Agentic AI Foundation](https://www.linuxfoundation.org/press/linux-foundation-announces-the-formation-of-the-agentic-ai-foundation), formed December 9, 2025, alongside MCP and goose. Anthropic's own CLAUDE.md is a Markdown file the model reads at the start of every session. So the barrier to "give an AI durable context" is now: create a Markdown file. Nothing else. There is a tell hiding in the vendor docs. Models lean on Markdown so hard that Anthropic ships a recipe to dial it back down. You wrap an instruction in ``, and the docs note that "removing markdown from your prompt can reduce the volume of markdown in the output." You fight to turn Markdown off. _When the people who built the model document how to suppress its default output, you have found the default._ ## Is your Markdown vault already a knowledge graph? Yes. The notes are nodes and the `[[links]]` are edges. Open any linked Markdown vault and you are looking at a graph. You have been keeping a knowledge graph without calling it one. This is not an AI invention. It is the 70-year-old Zettelkasten idea: atomic notes, connected by explicit links, where the connections carry the value. The split is clean: folders answer "where is this?"; links answer "what does this have to do with what?" That second question is the one an LLM cares about. A folder tells a model where a file sits. A link tells it what that file relates to. Folders are storage. Links are structure. Most of us already wrote the structure. We just never saw it. _A folder is an address. A link is a relationship. Only one of them survives contact with an LLM._ ## Does a personal vault really line up with Google's knowledge layer? Same shape. Not the same resolution. Here is the honest split. Google rebuilt search in 2012 around one idea: "things, not strings." Entities, and the relationships between them. Your notes are the things. Your `[[links]]` are the relationships. The web's open knowledge layer runs on the same primitive: schema.org, whose root type is literally `Thing`. Wikidata, the world's largest open-access knowledge graph, with over 120 million items. Google's own open Data Commons, nodes joined by directed edges. Nodes joined by edges, all the way down. That is the shape your vault already has. ![A two-column diagram: on the left, notes and links labelled node and edge; on the right, entities and typed relationships labelled Thing and bornIn, showing the same skeleton at different resolutions](/blog/your-notes-are-already-a-graph/images/fig2-same-skeleton.png) _Same skeleton, different resolution. A real knowledge graph types its edges. An Obsidian link does not, yet._ Now the honest half. Same shape is not the same resolution. A real knowledge graph types its edges: this node is a `Person`, that edge means `bornIn`. Obsidian links are untyped: every edge just says "related." A galaxy shows you the structure. It does not add the types. _Your vault has the skeleton of the web's knowledge layer. It does not yet have the labels on the bones._ ## Google just shipped Markdown, links, and a graph view Here is where the alignment stops being a metaphor. On June 12, 2026, Google Cloud published the Open Knowledge Format. It is vendor-neutral, tied to no cloud, model, or framework. And per the spec, "concepts link to each other with normal markdown links, turning the directory into a graph of relationships." A spec from Google now describes the exact directory I have been editing by hand for years. ![Two screenshots side by side: an OKF bundle as a directory of markdown files and its single-file HTML graph view, next to an Obsidian vault folder and its graph view, looking nearly identical](/blog/your-notes-are-already-a-graph/images/fig3-okf-vs-vault.png) _Left: Google's OKF bundle and its self-contained graph view. Right: an Obsidian vault and its graph. The convergence is structural, not cosmetic._ I did not borrow this format from Google. A directory of Markdown files with links is the obvious place to land: it is the lowest-common-denominator knowledge format that a human can read, an LLM can parse, and a graph can draw. > When the spec and the second brain converge on the same directory layout, the layout stops being a preference and starts being a standard. ## Is a 3D graph view actually useful, or just eye-candy? You do not need a plugin to see that your vault is a graph. Obsidian has shipped a graph view in the box for years. Open it and your notes are dots, your `[[links]]` are the lines between them. The structure was always there to look at. The catch is that the built-in view is flat. A 2D web of dots on a plane. It tells you the connections exist. It is harder to feel their depth, their clusters, the actual scale of what you have built. So I built a plugin called [Agentage Galaxy](obsidian://show-plugin?id=agentage-galaxy) that renders the same vault as a rotating 3D force-graph. Notes are nodes. `[[links]]` are edges. Top-level folders become colored clusters. Node size tracks the number of connections, so your hubs are the biggest stars. It is built on three.js and 3d-force-graph, and it mirrors the built-in graph, so there is nothing new to learn. The ownership receipts are simple. It is [MIT-licensed and open source on GitHub](https://github.com/agentage/obsidian-galaxy), and it makes zero network calls. Nothing uploads. It is live in the official Obsidian community directory, desktop-only, and installs in one click. Big vaults run heavier, since this is WebGL on the desktop. Be clear about what the view is. It is a diagnostic, not a brain. The graph does not retrieve anything or answer a question. What it gives you is the felt shape of your own thinking: the dense cluster you did not know was dense, the orphan note with no edges, the bridge node tying two topics together. That is real, and it is not the same as a model querying your knowledge. _A graph view is not a query engine. It is a mirror. Do not confuse seeing the shape of your knowledge with searching it._ ## What you can do with this in five minutes Open your vault and turn on the graph. Look for the orphans with no edges and the hubs with too many. That alone will tell you more about your knowledge than any tag system. Then notice what you are looking at: a directory of Markdown files, joined by links, readable by you and by every model you talk to. Google shipped a spec for it. The labs write their agent configs in it. You have probably been living in it for years. A graph view is not insight. It is the moment you can finally see that your notes were a graph all along. --- To see your own vault as a galaxy, [Agentage Galaxy](obsidian://show-plugin?id=agentage-galaxy) is one click in Obsidian: free, open source, and fully local. This piece was originally published on [Medium](https://medium.com/@vreshch). --- # Agentage Galaxy: see your Obsidian vault in 3D Published June 23, 2026. Source: https://agentage.io/blog/agentage-galaxy-3d-graph Your vault already knows its own shape. Every `[[link]]` you write is an edge, every note a node, every folder a cluster. The built-in graph view flattens all of that onto a plane. **Agentage Galaxy** lifts it into three dimensions, so the structure you have been building note by note finally looks like what it is: a galaxy. It is free, it is in the Obsidian community directory today, and it runs entirely on your device. ![A rotating 3D force-graph of an Obsidian vault](/blog/agentage-galaxy-3d-graph/images/demo.gif) ## What it is Agentage Galaxy is the built-in graph view, in 3D. The same data model and the same filters, rendered with [three.js](https://threejs.org) and auto-orbit: - **Auto-clustered by folder, zero config.** Every top-level folder gets its own color, so the shape of your vault pops on first open. Tags, attachments, and unresolved `[[links]]` each form their own group. - **Sized by connections.** Your hub notes are visibly the biggest stars. - **Nothing new to learn.** The same filters as the built-in graph - search, tags, attachments, existing-files-only, orphans - plus full force controls (center, repel, link, distance) and node size, link thickness, labels, and arrows. - **Fly through it.** Auto-orbit with a speed control. Left-drag to rotate, scroll to zoom toward the cursor, right-drag to pan, click a node to open that note. ![Agentage Galaxy running inside Obsidian, with the controls panel open](/blog/agentage-galaxy-3d-graph/images/in-obsidian.png) ## Install it One click from inside Obsidian: **[Open in Obsidian](obsidian://show-plugin?id=agentage-galaxy)** - or copy `obsidian://show-plugin?id=agentage-galaxy` into your browser. Or from the app: Settings -> Community plugins -> Browse -> search **"Agentage Galaxy"** -> Install -> Enable. Then click the brain icon in the left ribbon, and your vault starts rotating. ## 100% local Agentage Galaxy makes zero network calls. It is a pure, offline visualization of the Markdown you already own - no account, no upload, nothing leaves your machine. Desktop only, because it renders with WebGL. ## Part of Agentage Memory Agentage Galaxy shows you one vault. **[Agentage Memory](https://agentage.io)** is the layer underneath the idea: one set of plain-Markdown notes that every AI you use - Claude, ChatGPT, VS Code - can read and write over MCP, mirrored locally as files you own. > One memory. Every AI. Owned by you. The graph is the view. Memory is the thing every AI shares. Connect any client once at **[memory.agentage.io](https://memory.agentage.io)**, and your tools stop opening with a blank slate. Install [Agentage Galaxy](obsidian://show-plugin?id=agentage-galaxy), and see your vault the way it has always been shaped. --- # The Obsidian plugin is live - sync your vault to every AI Published June 21, 2026. Source: https://agentage.io/blog/obsidian-plugin-is-live Your Obsidian vault is already the best memory you keep. The problem was that the AIs you use could not read it - they opened blank every time while your context sat in files right next to them. Today that changes. **Agentage Sync is live in the [Obsidian community store](https://community.obsidian.md/plugins/agentage-memory).** It is a plugin that keeps your vault in sync with a private memory your AIs can read and write - so Claude, ChatGPT, Cursor, and any MCP client (any tool that speaks MCP, the open standard AIs use to connect to your data) work from the same Markdown you already own. [Install in Obsidian](obsidian://show-plugin?id=agentage-memory) ## Install Agentage Sync in Obsidian [Install in Obsidian](obsidian://show-plugin?id=agentage-memory) The button opens the plugin pane in your Obsidian desktop app. Desktop only today, mobile is on the way. You sign in once in the browser on first sync (OAuth, no API key), then pick or create a memory. Prefer to do it by hand? In Obsidian go to **Settings > Community plugins > Browse**, search **Agentage Sync**, then **Install** and **Enable**. Want the betas instead? Two more ways in: - **BRAT** - install [BRAT](https://github.com/TfTHacker/obsidian42-brat), then **Add beta plugin** > `agentage/obsidian-sync`. It keeps you auto-updated. - **Manual** - from the [latest release](https://github.com/agentage/obsidian-sync/releases/latest), drop `main.js`, `manifest.json`, and `styles.css` into your vault's `.obsidian/plugins/agentage-memory/` folder, then enable it. ## Set up Obsidian sync in four steps ### 1. Sign in to Agentage Open **Settings > Agentage Sync > Sign in to Agentage**. A browser window opens once to authorize the plugin (OAuth 2.1 with PKCE). There is no API key to copy, and the plugin stores no password - your token lives in Obsidian encrypted secret storage, never in your notes or config. ### 2. Choose a memory Click the status-bar dot (or run **Agentage Sync: Choose memory** from the command palette), then **Choose memory...** - pick an existing memory or create a new one. A fresh memory makes the cleanest first sync. ### 3. Sync now From the dot menu or the command palette, run **Agentage Sync: Sync now**. Your notes are committed and pushed. Sync is on demand today; background auto-sync is next. ### 4. Connect your AI The plugin gets your vault into the memory. The point is that every AI reads it. Point Claude, ChatGPT, Cursor, or any MCP client at the same memory over MCP, then ask it what is in your notes - it answers straight from your files. Sync a note from Obsidian, open a fresh Claude or ChatGPT chat, ask what it knows, and it reads the same memory. [See how to connect ->](/docs/connect) ![Diagram: your Obsidian vault two-way Git-syncs to a private memory that Claude, ChatGPT, and Cursor read and write over MCP](/blog/obsidian-plugin-is-live/images/flow.png) ## FAQ ### Do I need an account? Yes. You sign in to Agentage once, then pick an existing memory or create a new one. There is no API key to copy or manage. ### Where are my notes stored? As plain Markdown in your vault, and in a private Git repository - one per memory, hosted in the EU. The repo is the source of truth, readable as ordinary files, and exportable anytime. ### Is the plugin desktop only? Today, yes. Obsidian will not offer it on phones yet. Mobile is planned - the git engine already runs over Obsidian network layer, so what is left is verifying sign-in and first sync on real devices. ### Does the plugin send telemetry? No. There is no client-side telemetry, and a fresh or signed-out install makes no network requests at all. The plugin only reaches the network once you sign in or sync. ### Can I leave and take my notes? Yes. Your memory is a plain Git repo you can clone or export anytime, and your notes stay as Markdown on disk. Ownership is a property of the format, not a promise on a pricing page. ## Sync your Obsidian vault to every AI today Install Agentage Sync, sign in once, and run a first sync. The next AI you open reads your vault - because it is yours, not any one vendor's. [Install in Obsidian](obsidian://show-plugin?id=agentage-memory) One memory. Every AI. Owned by you. --- # Connect Claude to a markdown memory you own Published June 14, 2026. Source: https://agentage.io/blog/connect-claude-to-your-memory _First in a short series on giving every AI one memory you own. This one is Claude - Claude Code, Claude.ai, and Claude Desktop. ChatGPT and Cursor get their own posts._ Open a new Claude chat and it forgets everything from the last one. You re-paste the same context, re-explain the same project, every single time. This connects Claude to a memory it can read and write - so the next chat already knows, and so does every other AI you connect later. It takes about two minutes. One connector, the same for Claude Code, Claude.ai, and Claude Desktop. ## Connect now You need one URL. It is the same memory connector for every Claude surface: ```text https://memory.agentage.io/mcp ``` **Claude.ai (web) or Claude Desktop** - click [Add to Claude](https://claude.ai/customize/connectors?modal=add-custom-connector&connectorName=agentage%20Memory&connectorUrl=https%3A%2F%2Fmemory.agentage.io%2Fmcp). It opens Claude with the connector prefilled - just click Connect and sign in. (Or by hand: Settings > Connectors > Add custom connector, paste the URL.) **Claude Code** - run one command in your terminal: ```bash claude mcp add --transport http memory https://memory.agentage.io/mcp ``` The first time Claude uses the connector it opens a sign-in page in your browser to authorize it. That first sign-in creates your memory - there is no separate account step, and no API key to copy or paste back. Want to understand what you just connected? Keep reading. ## What Claude's built-in memory is, and where it stops Claude can carry some context for you now, and it helps. But it lives inside Claude. You cannot open it as a file, back it up, or point another tool at it - and it does not travel. Start a chat in another AI and that memory does not come with you. Agentage Memory is different. Your memory is a set of plain markdown files that you own. Claude reads and writes those files through the connector, you can open and export them anytime, and the same memory is readable by every other AI you connect. Nothing is locked inside one vendor. ## Connect your notes to Claude in four steps ### 1. Add the connector Use whichever Claude you already work in: - **Claude.ai or Claude Desktop:** click [Add to Claude](https://claude.ai/customize/connectors?modal=add-custom-connector&connectorName=agentage%20Memory&connectorUrl=https%3A%2F%2Fmemory.agentage.io%2Fmcp) - it opens with the connector prefilled, then click Connect. (Or by hand: Settings > Connectors > Add custom connector, paste `https://memory.agentage.io/mcp`.) - **Claude Code:** run `claude mcp add --transport http memory https://memory.agentage.io/mcp`. ### 2. Authorize in your browser The first time Claude reaches the memory, it opens a sign-in page. Approve it and the connection is live. There is no key to manage. ### 3. Ask Claude to save something Tell Claude to remember something worth keeping. For example: ```text Save to memory: I'm building a recipe app in Next.js, and I prefer TypeScript with no semicolons. ``` Claude writes it as a markdown note in your memory. ### 4. Open a new chat and recall it This is the test that proves it. Open a fresh chat with none of the earlier context and ask: ```text What do you remember about my project? ``` Claude searches your memory, finds the note, and answers with your project and preferences. Write, new chat, recall - if that round-trip works, your memory is connected. ## How it compares There are good memory tools out there. They tend to split into two camps, and each gives up something: | Tool | Where your memory lives | Setup | Same memory in other AIs | | ------------------------ | ------------------------------ | --------------------- | ------------------------ | | Claude's built-in memory | Inside Claude, no export | None | No | | Mem0 | A hosted cloud database | Hosted, quick | Some clients | | Basic Memory | Plain markdown, run locally | Local, manual config | Yes, via MCP | | Agentage Memory | Plain markdown you own, hosted | Paste one URL, ~2 min | Yes, one connector | Credit where it is due: Mem0's hosted setup is genuinely smooth, and Basic Memory really does hand you your markdown files. The catch is what each one trades away - Mem0 keeps your memory in a database you cannot open, and Basic Memory asks you to run and wire it up yourself. Agentage Memory is the row that is hosted and convenient, plain files you own, and shared across every AI - without making you pick one or the other. ## Already use Obsidian? Your Agentage memory is plain markdown with YAML frontmatter - the same shape an Obsidian vault uses. The notes Claude writes look and read like notes you would write yourself, and because you can export your memory anytime, nothing traps them in a format only a machine can read. If you live in markdown, this will feel familiar. ## One memory across Claude, ChatGPT and Cursor This post connects Claude. The bigger idea is one memory that every AI shares. Save a decision while working with Claude today, and the AI you open tomorrow reads it from the same files - because the memory is yours, not any one vendor's. Connecting ChatGPT and Cursor is the next post in this series. ## Why not just use Claude's own memory? Honest answer: if you only use Claude, and you are fine with your memory living inside Claude where you cannot export it or point other tools at it, the built-in memory may be enough. Use it. Agentage Memory is for the other case - you want your memory as plain files you own and can take anywhere, and you want the same memory to follow you across more than one AI. That is the whole point. One honesty note on search: memory search matches the words you actually wrote, like a keyword search over your files. It is literal, not a fuzzy meaning-match, so search for the terms you would use. ## FAQ ### Do I need an API key? No. You connect with one URL or one command, and the first time Claude uses the memory it opens a browser sign-in to authorize. There is no key to copy, paste back, or rotate, and nothing for you to store. ### Where are my notes stored? Your memory is plain markdown files that belong to you, stored on EU infrastructure and exportable anytime. They are the source of truth, readable as ordinary files - nothing is locked inside one vendor. ### Does this replace Claude's built-in memory? It is a separate, file-based memory you own and can take elsewhere. Claude's built-in memory stays inside Claude. Use Agentage Memory when you want plain files and the same memory across more than one AI. ### Is the search smart or keyword based? It is keyword based. Search matches the actual words in your notes, not the meaning behind them, so use the terms you would expect to find. It keeps results predictable when you go looking for something. ### How long does setup take? About two minutes. You add the connector once - paste the URL in Claude.ai or Claude Desktop, or run one command in Claude Code - then authorize in your browser the first time Claude uses it. ## Connect your notes to Claude now One connector, about two minutes, and the next chat already knows. - **Claude.ai or Claude Desktop:** click [Add to Claude](https://claude.ai/customize/connectors?modal=add-custom-connector&connectorName=agentage%20Memory&connectorUrl=https%3A%2F%2Fmemory.agentage.io%2Fmcp) - it opens with the connector prefilled, then click Connect. (Or by hand: Settings > Connectors > Add custom connector, paste `https://memory.agentage.io/mcp`.) - **Claude Code:** `claude mcp add --transport http memory https://memory.agentage.io/mcp` Write a note, open a new chat, ask Claude what it remembers. When it answers, your memory is connected - and it is yours to keep. One memory. Every AI. Owned by you. --- # The MCP endpoint is live - connect your AI today Published June 7, 2026. Source: https://agentage.io/blog/mcp-endpoint-is-live A week ago we opened the waitlist. Today the MCP endpoint is live in production. **The Agentage Memory MCP endpoint is up at https://memory.agentage.io/mcp** - live in production today. Point an MCP client at it, sign in once in the browser, and that client is reading and writing your memory - the same memory every other client you connect will see. Connect once, and your tools stop opening with a blank slate. ## Connect it Pick your client. They all end at the same URL. **Claude Code** - one line in the terminal: ```bash claude mcp add --transport http memory https://memory.agentage.io/mcp ``` Then run `/mcp` and complete the sign-in. **Claude (claude.ai and Desktop)** - open [Settings > Connectors](https://claude.ai/customize/connectors?modal=add-custom-connector) > Add custom connector, paste the endpoint, sign in when prompted. **VS Code** - add this to `.vscode/mcp.json` (or use the `MCP: Add Server` command): ```json { "servers": { "agentage-memory": { "type": "http", "url": "https://memory.agentage.io/mcp" } } } ``` The browser opens for sign-in on first use. **ChatGPT** - open [Settings > Connectors](https://chatgpt.com/#settings/Connectors) > Add custom connector on paid plans (enable Developer mode under Advanced if you do not see it), paste the endpoint, sign in. Anything else that speaks MCP over Streamable HTTP can point at the same URL. If your tool can speak MCP, it can speak to your memory. There is nothing of ours to install. Full per-client steps are on the [docs page](/docs). ## What happens on first use The first call opens your browser. Sign in with a magic link - under the hood that is OAuth 2.1 with PKCE and dynamic client registration - meaning no API key to copy, nothing to paste into a config, and a token scoped to your memory only. Discovery metadata is published at the usual well-known endpoints, so a client we have never heard of can find the authorization server and wire itself up without asking us. Your memory is bound to your account - one memory per user, zero IDs to manage. ## The six tools Once connected, your AI has six tools: - **memory\_\_search** - find notes by keyword across the whole memory. - **memory\_\_read** - read a note by path. - **memory\_\_write** - create or replace a note. - **memory\_\_edit** - apply a targeted edit to a note. - **memory\_\_list** - list notes, optionally under a folder. - **memory\_\_delete** - remove a note. One honest detail: search is literal keyword matching, not semantic - it finds the words you actually stored, fast and predictable. The notes themselves are plain markdown with YAML frontmatter, EU-resident by architecture. Ownership is not a roadmap item: you can export the whole memory today. Zero lock-in is a property of the format, not a line on a pricing page. The alternative - a database you cannot see, in a format you cannot take - is exactly what we are not building. If we disappeared tomorrow, your memory would not. ## Connect today If you run more than one AI and you are tired of being the memory between them, you do not have to wait anymore. Point one client at https://memory.agentage.io/mcp, sign in, and write your first note. It takes about two minutes, and the next tool you connect will already know it. One memory. Every AI. Owned by you. --- # Agentage Memory is open - join the waitlist Published May 31, 2026. Source: https://agentage.io/blog/agentage-memory-is-open If you work with AI all day, you probably run four or five tools to do it: Claude for one thing, ChatGPT for another, Cursor in the editor, Claude Code in the terminal. Each one opens with a blank slate. None of them knows what the others already learned about you, your project, or the decision you made last week. So you became the memory. Every time you switch tools you re-paste the architecture, the constraints, the context - fifteen minutes here, twenty there, several times a day, forever. And when a vendor quietly deprecates its built-in memory or wipes it on the next update, the context you spent weeks shaping is simply gone. Today we are opening the waitlist for the thing we are building to end that: **Agentage Memory**. ## One memory, every AI Agentage Memory is a single shared memory your AI tools read and write through one MCP endpoint. You connect once - Claude, ChatGPT, Cursor, Claude Code - and from then on they all see the same notes. No per-tool setup, no copy-paste, no plumbing. Connect in minutes and every AI is on the same page. The canonical copy lives in your own per-tenant store, hosted in the EU. It mirrors to plain markdown files on your machine - files you can open in Obsidian, edit by hand, grep, version, and export anytime. The memory is yours, in a format that outlives any one tool. ## Why build another one There is no shortage of "AI memory" products. Most fall into two camps, and both have a catch. Closed-cloud memory keeps your notes in a database you cannot see and do not own. Vendor-built memory - ChatGPT's, Claude's, Cursor's - lives inside one tool and disappears the moment that tool changes its mind. We are betting on the opposite of both. Four things, together, that nobody else offers as one package: - **Files you own** - plain markdown on your disk, exportable anytime, zero lock-in. - **Every AI** - cross-vendor by default, including a ChatGPT connector, not a walled garden. - **Your own store** - a per-tenant copy of your data, not a shared cloud blob. - **EU-resident** - your data stays in your region by design. All at a flat, low price, with no surprise jump from a few dollars to a few hundred. The winner of AI memory will not be a closed cloud database - it will be the open, files-first memory every tool agrees to share. ## It is early - on purpose Let us be honest about where this is: it is day one. Agentage Memory is pre-release - not in any plugin or community store yet, and not generally available. We are building it in public, shipping in the open, and every integration will land as its own story you can follow: the MCP endpoint, the sync engine, the Obsidian plugin, the CLI. The waitlist is how we bring people in deliberately, in order, while it is still rough enough that your feedback actually shapes what it becomes. ## Join the waitlist If you run more than one AI and you are tired of being the memory between them, get on the list. We will reach out as we open spots - and you will be early to a memory that is finally yours. One memory. Every AI. Owned by you.