Solo mode
In Permissions, the MCP writes switch applies to all MCP keys. Turn it off for a read-only first run. Individual key write settings are not used in Solo mode.
GETTING STARTED
Start with one client, one brand, and one useful task. Build the context once, then retrieve it from a connected AI agent.
You need access to an Agency Memory workspace and an MCP-capable agent. Owners and admins manage the app. If an admin has already issued you an MCP key, go straight to connecting your agent. This guide begins with an existing workspace; it does not create an account.
A client is the organization you work with. A brand is the business, product, or property with its own positioning, voice, and accounts.
In the app, open Clients → New client. Leave Create a first brand selected for a typical single-brand client. Give the brand a different name if needed. For a parent company whose brands you will add later, clear that option.
Start with one brand you know well. You can add more when you have checked the first workflow.
Open the brand and find Brand Knowledge → Add knowledge. Write a short brief in the editor, then choose Save knowledge. You can use plain text or Markdown.
Use the full brand brief template if you need more structure. Brand Knowledge is one maintained document per brand; saving a change creates a new version.
Add a Playbook when you have a recurring process to document. Use Platforms for account and asset identifiers, relationships, and brand assignments. Registering those details does not connect the external provider.
Every active MCP key can read the whole workspace, including every client and brand. Choose a connection you trust with that scope.
In Permissions, the MCP writes switch applies to all MCP keys. Turn it off for a read-only first run. Individual key write settings are not used in Solo mode.
In Agent access, each MCP key has its own Can edit memory setting. Leave it off for a read-only first run, then enable it only for connections that should update context.
Owners and admins can still edit through the app. Allowed memory changes apply directly; there is no approval queue. A write setting controls Agency Memory records, not advertising permissions or spending.
Open Agent access → Create MCP Key. Enter the user’s name, an optional email, and a label that identifies the connection, such as “Alex — laptop agent.” In Agency mode, check whether Can edit memory should be enabled before creating it.
Copy the full key when it is shown. The saved key record keeps a hash and an identifying prefix; it cannot reveal the original secret later. Store it privately. Create a separate key for each person or connection so you can identify and revoke it independently.
If you leave the screen and no longer have the secret, create a new key for the connection and revoke the superseded key after updating your agent.
In Agent access, choose Claude Desktop, Claude Code, Codex, Cursor, or Other. Select an active MCP key created during your current visit, then choose Copy prompt. The prompt includes the full key: share it only with the agent you intend to connect.
The app supplies the MCP URL and the client-specific instructions. Follow them, preserve any existing server configuration, and restart or reload your agent when prompted.
Compatibility depends on the client supporting the required transport and authorization header. A web chat that only supports OAuth is not interchangeable with this connection.
Use the Agency Memory MCP to list my clients and brands. Report the result or the exact connection error. Do not create or change any records.
An empty list can be a valid result for an empty workspace. Saved settings or a discovered tool list alone do not prove that a request succeeded.
Once the connection works, ask your agent to retrieve the brand you prepared and draft something you already know how to review.
Retrieve the current Brand Knowledge for [client / brand] from Agency Memory. Summarize the audience, offer, voice rules, and supporting evidence you will use. Flag missing information. Draft three campaign message directions using that context. Return drafts for review; do not update memory or publish anything.
Check the brand, offer, claims, and tone against the brief you saved. Keep the setup focused on one useful output before adding more clients or tools.
Your first milestone is a successful context retrieval used in real work. The next is being able to repeat it in a fresh conversation without rebuilding the brief.
Email [email protected] with the agent name and a redacted error message. Never include your MCP key.
ONE CLIENT. ONE USEFUL TASK.
Keep the brief current, check the output, and repeat the workflow in a new conversation.