Should You Control AI Marketing Agents from Claude Instead of a Dashboard?
Run agents from Claude when you want to trigger one agent's work from inside a chat or a workflow you already have open, and run them from the dashboard when the output needs review, approval, or a client's sign-off before it ships. Orbitable supports both paths from the same fleet: an MCP server that exposes all 50 specialist agents as tools to Claude and any other MCP client, and a dashboard built around the Dock, a review surface where drafts get marked up, sent back, and approved into a Library. Neither path is the "right" one in isolation. The choice depends on whether the next step after the agent finishes is more work, or a decision.
What Calling an Agent From Claude Actually Does
The MCP server exposes each of the 50 agents, across the 10 specialist squads, as a callable tool. Ask Claude to run the seo-strategist or the case-study-builder, and it dispatches to that agent using the same shared world model every agent draws on: your ICP, brand voice, products, competitors, and whatever you've uploaded to the knowledge base. That shared context is the actual value here, not the novelty of typing into Claude instead of a web app. An agent triggered from Claude still knows the same customer, the same positioning, the same prior work, as one triggered from the dashboard.
What you get back is whatever that agent produces, inside the Claude conversation. There's no built-in markup, no pinning comments to a specific paragraph, no routing to a second person for approval. It's a direct request-and-response loop, which is exactly what you want when you're iterating on something yourself and don't need anyone else to weigh in before it goes anywhere.
What the Dashboard and the Dock Add
The dashboard is where the Dispatcher, the orchestrator that coordinates the 50 agents (it is the orchestrator, not a 51st specialist), sits alongside every playbook, schedule, and document template. The Dock is the review layer on top of that: a client or internal stakeholder sees what's waiting, marks it up with pins directly on the draft, sends it back for changes, and approves it into a Library once it's done. Requests in the Dock carry SLA targets, so a stakeholder can see not just what's pending but when it's due.
None of that exists inside a Claude conversation. If the deliverable needs a second set of eyes, a client's approval, or a paper trail of what was requested and when, it needs to pass through the Dock. Claude can produce the draft. It can't run the approval loop.
Autopilot Only Proposes, It Never Approves From Claude
Autopilot proposes a weekly plan: one focus, 3 to 5 missions, each with a why-now, a carry-over list of what didn't finish, and an explicit NOT-doing list of what's being deliberately skipped. It is available on paid plans, and it proposes only. Approval always happens in the Dock, regardless of whether the plan itself was generated by an agent you triggered from Claude or from the dashboard. This is a structural point, not a preference: there is no path for Claude to approve a weekly plan on your behalf. If you're using Autopilot to run your week, you're going to open the Dock at least once, no matter how much of the drafting happens elsewhere.
Credits Work the Same Way in Both Places
Credits meter work, not the interface you used to request it. An agent call costs the same whether it was triggered from Claude via MCP or from a mission inside the dashboard. Reviewing work never costs credits either way, so opening the Dock to mark up a draft, request changes, or approve it into a Library doesn't touch your balance. When credits run out, work pauses until a prepaid pack tops it up or the month resets. There's no overage billing, and no difference in how that limit applies depending on where the request came from.
Comparing the Two Paths Directly
When You Actually Need Both
Most working setups end up using both paths for different jobs, not one instead of the other. A founder running solo might trigger the case-study-builder from Claude while drafting a proposal, then never touch the Dock because nobody else needs to sign off. A team with a client on the Dock will still have someone open the dashboard to run a playbook or check Autopilot's weekly plan, because that's where the SLA-tracked requests and the approval trail live. The MCP server and the dashboard aren't competing products. They're two entry points into the same 50-agent fleet, sharing the same world model, so work started in one place doesn't have to be re-explained in the other.
FAQ
Can I approve a draft from inside Claude?
No. Approval always happens in the Dock, regardless of where the draft was generated. Claude can produce or revise the work, but marking it approved into a Library, or sending it back with pins, is a Dock action only.
Does triggering an agent from Claude cost more credits than using the dashboard?
No. Credits meter the work an agent does, not the interface used to request it, so a call via MCP costs the same as the equivalent call from the dashboard. Reviewing work never costs credits in either place.
Can a client see work that was generated through Claude?
Only once it reaches the Dock. The Dock is the client-facing review surface, included on Agency and Enterprise and available as a $49/mo add-on on Founder and Team; a client seat is Dock-only and never sees internal team notes.
Does Autopilot work if I only use Claude?
Autopilot will still propose a weekly plan, one focus with 3 to 5 missions, a carry-over list, and a NOT-doing list, but approval always happens in the Dock. You cannot approve the plan from Claude itself.
Which plans include the MCP server?
The MCP server exposes every agent as a tool to Claude and any other MCP client, driving the fleet from outside the app itself. The Dock, by contrast, is included on Agency and Enterprise, and available as a $49/mo add-on on Founder and Team.