Know what your agents can access.

Agency Memory gives your team and connected AI agents a shared source of client and brand context. Organize clients and brands, maintain Brand Knowledge, write repeatable Playbooks, and record the platform accounts and assets each brand uses. Agents retrieve that context through MCP, the Model Context Protocol.

A document can be a useful brief. Agency Memory adds a client and brand structure, access through MCP, write controls, and version history. A connected agent can retrieve the current brand knowledge for a new task instead of relying on what was pasted into an earlier conversation. You still need to ask the agent to retrieve the right context and review its output.

The app provides connection instructions for Claude Desktop, Claude Code, Codex, Cursor, and other compatible MCP clients. The connection uses Streamable HTTP with an MCP key sent as a bearer token. Claude Desktop uses a local bridge. A client that only offers an OAuth connection or cannot send the required authorization header may need a different setup. A regular browser chat is not automatically compatible.

No. Every active MCP key can read all clients, brands, and records in its workspace. The permission controls decide whether it can also change memory; they do not limit which brands it can read. Issue keys only to people and agents you trust with that workspace’s contents.

In Solo mode, the MCP writes switch in Permissions controls writes for every MCP key. In Agency mode, each key has its own Can edit memory setting in Agent access. Owners and admins can edit through the app. Changes to an MCP key’s effective write permission apply on its next request.

Agency Memory supplies context. Your separately configured advertising, analytics, and creative tools perform external work. Registering an account in Platforms records its identifiers, relationships, and brand assignment; it does not authenticate with the provider or grant permission to publish, change budgets, or spend.

Brand Knowledge is the brand’s maintained written brief: positioning, audiences, voice, offers, and guidance. Playbooks contain repeatable instructions for a particular kind of work. Platforms is the structured inventory of accounts and assets, their links, and their brand assignments. Keep passwords and provider API secrets in your credential system.

The admin app is for workspace owners and admins. Other teammates can use an MCP key issued by an admin to connect their agent. A named key identifies the person or connection behind the work. It does not give that person an admin login or a separate private area inside the workspace.

Changes permitted by the caller’s write access apply directly; Agency Memory does not put them into an approval queue. Version history and audit records help you trace changes afterward. If your workflow requires human review, begin with a read-only key and review proposed changes before enabling writes. Instructions in a Playbook do not replace permission controls.

You can revoke a key from Agent access. Requests using a revoked or expired key are refused. The full key is shown when it is created and cannot be retrieved later from the saved key record. If you lose it, create a new key, update your agent’s connection, and revoke the old one.

The connected agent receives the information it requests within the workspace access granted by its MCP key. Its provider’s terms and data settings govern how it handles that information. Choose agents appropriate for your clients’ confidentiality requirements, and review data and agent access before sharing workspace content.

Email support@agencymemory.ai for setup help, access questions, or privacy and deletion requests. Archiving records preserves their history and is different from permanent deletion. Do not send MCP keys, passwords, or confidential client material in a support message.