Docs

One MCP memory server for all your agents

Give your agents a shared memory. Every agent you run starts cold: Claude doesn't know what Cursor figured out an hour ago, and your own agent doesn't know either. agentics.you is one MCP memory server that all of them connect to. What one agent remembers, the others can recall.

Start free. Shared memory is free forever, and you don't need a card to start.

One address for every client

Every client uses the same address:

Text

https://agentics.you/mcp

That includes Claude, ChatGPT, Cursor, Grok, Hermes, OpenClaw, Claude Code, Codex, Gemini CLI, VS Code, Windsurf, and any other MCP client. Connecting takes about two minutes and needs no code. You sign in once and pick which agent this connection is.

Cursor (~/.cursor/mcp.json):

MCP

{ "mcpServers": { "agentics": { "url": "https://agentics.you/mcp" } } }

Claude Code:

Text

claude mcp add --transport http agentics https://agentics.you/mcp

Codex:

Text

codex mcp add agentics --url https://agentics.you/mcp
codex mcp login agentics

Clients that only run a local process use the published package @agenticstechnologies/mcp-server. It forwards stdio to the same address and signs in with an agent key (agk_mcp_…) taken from AGENTICS_AGENT_KEY:

MCP

{ "mcpServers": { "agentics": { "command": "npx", "args": ["-y", "@agenticstechnologies/mcp-server"], "env": { "AGENTICS_AGENT_KEY": "agk_mcp_..." } } } }

Keep the key in the host's environment, not in a file you commit.

Step-by-step pages: Cursor · Claude Code · Claude Desktop and claude.ai · ChatGPT · Codex · Gemini CLI · VS Code · Windsurf · Grok · any MCP client.

Check it worked: remember, then recall

Ask your agent:

Paste this into your agent

Call the remember tool with content "The customer prefers email over phone." Then call recall with the query "how does the customer like to be contacted?" and tell me the receipt_hash.

remember answers with saved: true, a receipt_hash and a verify_url. recall returns that memory with the same hash. Now connect a second client and run only the recall. It finds the memory the first agent wrote. That's shared memory for agents.

Memory you can check

Every memory write and every rule decision is on the record as a receipt. A receipt names the workspace, the agent, the action and the rule decision, and holds a SHA-256 digest of the content. Each receipt points to the one written just before it. Receipts are sealed into a batch every 30 minutes, and the batch is anchored to a public record.

Anyone can check a receipt at /verify/ with no account. The check never shows the memory's text. It shows the record wasn't changed after it was written. It doesn't prove the memory is true. How receipts work.

Your agents know their limits

Rules decide what an agent can do on its own and what waits for your yes. Anything that waits shows up under Needs you. Rules, approvals and Pause all are on every plan, including Free.

Prefer HTTP to MCP?

The same memory is available over a plain API in four calls: create a key, write, check the job, search. See the agent memory API. SDKs: pip install agentics-you and npm install @agenticstechnologies/sdk.

Pricing

Free, $0, forever: 10,000 actions a month, 10,000 stored memories, 2 people, unlimited agents. Recall is never counted.
Pro, $29/mo: no stored-memory cap, 100,000 actions a month, up to 3 people, plus exports, webhooks and invoices from receipts.
Team, $199/mo: 1,000,000 actions a month, unlimited people, plus the audit log view and export, roles and team budgets.
Enterprise, from $1,500/mo (billed annually): SSO (SAML/OIDC), custom retention and DPA, data residency, named contact and SLA.

We price on actions, never per agent. Full pricing.

Start free