Picmim works with external AI agents such as Claude, Cursor, and ChatGPT through the Model Context Protocol (MCP). Once connected, an agent can read your accounts, posts, calendar, media library, and analytics, and, if you allow it, create and schedule content on your behalf.
What you can do
A connected AI agent acts like a teammate with limited access. It can look up your connected accounts, find and draft posts, spot gaps in your calendar, pull media from your library, inspect raw analytics, and help review generated content Plans. You stay in control because every action the agent can take is one your workspace granted it.
- Read what is in your workspace, including accounts, posts, calendar, and analytics.
- Draft and schedule posts when the connection has write access.
- Suggest posting times based on your calendar gaps.
- Summarize analytics and explain performance in plain language.
- Review, rewrite, reschedule, and request visuals for generated Plan items when granted Plan access.
Connect with OAuth
OAuth is the preferred way to connect an AI agent to Picmim. It never exposes a long-lived credential to the client, and you approve the exact workspaces it can reach.
- Start the connection from inside the AI client, for example by adding Picmim as a connector.
- Sign in to Picmim when the browser opens.
- Choose which workspaces to share with the agent.
- Approve the connection to finish.
Tip: New OAuth connections start with every supported capability and automatic execution enabled. Workspace owners can later disable individual capabilities or require approval for every write action.
Capabilities and approval
Capabilities decide what an agent is allowed to do once it is connected, and approval mode decides whether it can act immediately.
- By default, a new connection can use every supported capability and executes actions automatically.
- A workspace owner can disable individual capabilities the agent should not have.
- A workspace owner can switch a connection to approval-required mode, so every write action waits for approval in Picmim before it runs.
- Every request still respects the connecting member's workspace role, regardless of capability settings.
Workspace credits and roles
Work an agent performs on your behalf is not free-standing: it draws on the same workspace resources as the app.
- AI work such as generating plans, captions, or visuals uses workspace credits at the same rates as in-app usage.
- Workspace roles and credit limits still apply, so an agent can only do what its connecting member is allowed to do.
Personal tokens
Personal tokens are an alternative to OAuth for clients and automation environments that provide a protected secret store, such as a server-side integration. Avoid them in any client that cannot keep a secret safe.
- Sign in to Picmim and open the target workspace.
- Open Workspace settings, then select MCP.
- Create a named token and choose a scope: read or write.
- Copy the token immediately. You will not be able to see it again.
Warning: Treat a personal token like a password. Anyone who has it can access your workspace with the scope you chose.
Connect an agent
Every supported AI client has its own connection steps, covering ChatGPT, Grok, Muse Code, Claude, Cursor, Gemini CLI, and VS Code with GitHub Copilot. See the AI client setup page for exact instructions for your client.
Security and control
Both OAuth connections and personal tokens are designed to be safe to create, share within your team, and revoke.
- Members can rotate or revoke their own personal tokens at any time.
- Workspace administrators can review and revoke any workspace credential, OAuth connection or personal token, at any time.
- Only a hashed version of a personal token is stored, so it cannot be read back.
- Every request re-checks that the connecting member is still part of the workspace.
- Set an expiry on personal tokens so long-lived credentials expire on their own.
Warning: If a token or connection may have leaked, revoke it immediately from Workspace settings, then MCP, and create a new one.
If it is not working
Most connection problems come from a small number of causes.
- OAuth loop or consent errors: disconnect the connector in the AI client and reconnect from scratch.
- Expired or revoked credential: reconnect with OAuth, or generate a new personal token from Workspace settings, then MCP.
- Wrong scope: creating or editing content needs write access, either from capabilities or from a write-scoped personal token.
- Action waiting instead of running: check whether the connection has been switched to approval-required mode.
- Local HTTPS certificate: when testing against a local Picmim, your agent may need to trust the local certificate or skip certificate verification.
Tip: If the agent connects but cannot do something, check its capabilities or token scope first.