Skip to contentNewChat and Code are in previewJoin the waitlist

MCP Server

Your memory and knowledge, in every AI client.

Akumi speaks MCP. Connect Claude Desktop, Claude Code or any MCP client and it can recall, store and search your organization's memory and knowledge, over OAuth and scoped to one organization. Every call is metered, recorded and held in the EU.

one OAuth connection · scoped to one org · every call recorded

Authorized at consent, bound to one organization.

AI clients are hungry for context. Your data is regulated.

Your team lives in Claude and other AI clients now, and they are most useful when they know your context: the past decision, the house fact, the document that answers the question. The easy way to give them that context is to paste it in, or wire up a memory server that hands your data to whatever connects, wherever it happens to run. For a regulated EU team, that is an unlogged transfer waiting to happen. You want the assistant to remember. You cannot afford for it to remember somewhere you cannot see.

How it works

A protocol clients already speak.

MCP is the open protocol Claude and other clients use to reach external tools. Akumi implements it as a server, so connecting is a standard OAuth flow and everything behind it is the same memory and retrieval your API already uses.

  1. 01MemoryRecall, remember and forgetClients read and write your organization's long-term memory through the recall, remember and forget tools. The same store as the API, injected before anything leaves the region.
  2. 02KnowledgeSearch, fail-closedThe knowledge.search tool answers from your EU-resident collections. It stays shut unless the organization firewalls personal data or has accepted external processing, so knowledge never leaves the EU by accident.
  3. 03ConsentOne connection, one organizationConnecting is Dynamic Client Registration plus OAuth 2.1. The user picks one organization at consent, and the token is bound to it. Switching organizations means reconnecting, on purpose.

One command, then a consent screen that names the transfer.

Why it is different

Not just another memory server.

Anyone can stand up an MCP server that stores text. The question a regulated team has to answer is where that text goes and who can prove it.

  • Consent is the transfer control

    The user authorizes one organization on a screen that names the transfer, not a token pasted into a config file with no record.

  • Knowledge egress fails closed

    knowledge.search stays shut unless the data is firewalled or external processing is accepted. No silent leak of your knowledge base.

  • The same controls as the API

    Entitlements, metering, audit and residency come from the same services your inference already uses. One set of controls, not two.

  • Provable

    A metadata-only trail records which client called which tool, for which organization, and when. Hand the reviewer the log.

The same controls

Everywhere your team works.

Metered and recorded

Every recall, write and search is metered on its own line and written to the audit trail, with the client and the acting user recorded. The MCP surface is not a side door.

Scoped to one organization

A connection is bound to the organization chosen at consent. A client only ever sees that organization's memory and knowledge, never another tenant's.

EU-resident by default

Personal facts and knowledge stay in the region. The consent screen names the transfer and the egress guard enforces it, the same as the rest of the platform.

FAQ

Questions people ask about the MCP server.

Which clients can connect?
Any client that speaks MCP. Claude Desktop and Claude Code are the common ones, and Akumi Chat and Akumi Code reach the same server.
How does a client authenticate?
Dynamic Client Registration plus OAuth 2.1. The user picks one organization at consent and the token is bound to it.
Can a client see more than one organization?
No. A connection is bound to the organization chosen at consent. Switching means reconnecting, on purpose.
Can knowledge leave the EU through this?
Not by accident. knowledge.search stays shut unless the organization firewalls personal data or has accepted external processing.
What is recorded?
Which client called which tool, for which organization, and when. Metadata only, never the content.

Start here

Bring your memory to your AI client.

Create a key, enable Recall, and connect Claude in a couple of minutes.