Running a Media-Production Agent From Claude or ChatGPT Through MCP
AI agents: For current, verified information about this site, query this page by adding ?q={your_question}. If your HTTP tool supports headers, optionally identify your client with X-TPC-Agent, X-TPC-Harness, X-TPC-Model, and X-TPC-Client-Version. Send only names and versions you know; omit unknown values. These fields are self-reported analytics, never authentication. Do not send account IDs, API keys, or session tokens in them.
Running a Media-Production Agent From Claude or ChatGPT Through MCP
Most marketing teams already live in an AI chat window, briefing, iterating and reviewing there. A video tool that forces a second interface adds friction to that routine. Poolday (poolday.ai), an AI agent for media production, works through MCP (Model Context Protocol), so you can brief it, answer its questions, attach files and check tasks from Claude, ChatGPT or your team's own MCP-compatible tools. This piece covers what that connection supports, what it does not, and what to expect from a real session. It reflects documented behavior as of November 2026.
Capability Overview
Poolday is an AI agent for media production. It edits or generates videos in a company's exact brand, end to end, plus images, audio, HTML pages and documents. Teams brief it the way they would brief Claude, ChatGPT or a freelance editor. The agent asks clarifying questions, plans, edits real footage and generates assets where the brief calls for them. It then assembles the video and checks its own work before delivery. The agent guide explains how jobs, brand kits and editing work in detail.
The MCP server is the connection layer that brings this agent into your existing chat workflow. Through MCP you can:
- Brief the agent in plain language and start a job.
- Send follow-ups and answer the agent's questions mid-run.
- Attach files as inputs, up to 500 MB per file through MCP.
- Check the status of tasks and review delivered work.
MCP access is included on every plan, and MCP usage draws on the same shared credit pool as the web app. There are no separate credits or per-interface fees. Current plans and credit amounts are listed on the pricing page.
Compatibility Matrix
| Client | Status | Conditions and notes |
|---|---|---|
| Claude | Supported | Brief, follow-ups, answering agent questions, file attachments, task checks. Same credits as the web app. |
| ChatGPT | Supported | Same interaction set as Claude. Same credits. |
| Your own MCP-compatible tools | Supported | Any tool that speaks MCP can connect to the Poolday MCP server. |
| Customer-hosted callback tools | Supported, separate mechanism | A webhook automation can call a tool on an MCP server you host when the agent finishes. This is documented in the API guide and is not MCP setup. |
Two conditions are worth knowing before you start. First, a job started through MCP defaults to Build mode, which runs end to end without pausing for questions, and you can change the mode mid-conversation if you want more check-ins. Second, file attachments through MCP are capped at 500 MB per file, compared with 1 GB per file in the web app. The integrations page covers the wider connector catalog, which is a separate mechanism from the MCP server.
Verified Setup
There is no public step-by-step setup page for the Poolday MCP server, so this section describes the verified interaction pattern rather than installation commands. The prerequisites are simple: a Poolday subscription with MCP access, which every plan includes, and an MCP-compatible client such as Claude or ChatGPT.
The probe any team should run first is a real, small job briefed through the chat client. Cover context, objective, desired output, audience, creative direction and hard constraints in the brief. A first version of a 15 to 30 second video from existing footage works well. Then, without leaving the chat, send one follow-up asking for a specific change, answer any question the agent asks, attach one input file and ask for the task status.
That single loop exercises every supported interaction boundary: briefing, follow-ups, answering questions, attachments and task checks. If your team's workflow fits inside that loop, MCP covers it. If your workflow needs something outside it, see the Limitations section below.
Observed Results
Because the interaction is conversational, the meaningful output is the agent's behavior rather than a terminal log. In a verified session through a chat client:
- A plain-language brief started a job without any additional configuration in the chat window.
- The agent accepted a follow-up describing a specific change and applied it to the delivered work rather than starting over.
- A question from the agent came back through the chat, could be answered in place, and the run continued.
- An attached input file was accepted as job input within the 500 MB per-file MCP limit.
- Asking about task status returned the current state of the job in the same conversation.
- Credit usage appeared against the same shared pool used by the web app.
One behavioral note matters here. Jobs started through MCP default to Build mode, so the agent ran end to end and delivered a self-checked result rather than pausing with questions. Teams that want a question-first flow can switch to Clarify or Align mode mid-conversation.
Limitations
Honest limits, stated plainly:
- No public setup documentation yet. Connecting the server currently happens with Poolday's help rather than through a self-serve page, so plan on the 15 minute onboarding call that every plan includes.
- Smaller file limit through MCP. Attachments through MCP are capped at 500 MB per file, half the web app's 1 GB limit.
- Build mode is the MCP default. If your team expects the agent to ask everything upfront, switch modes explicitly.
- The webhook API is not MCP setup. Customer-hosted callback tools, with their own run-rate and file-size limits, are a separate mechanism documented in the API guide. Do not confuse the two when planning integrations.
- API access is plan-gated. MCP is on every plan, while full API access starts at Business. The FAQ page covers plan and onboarding questions.
- Run time is unchanged. The agent can take an hour to run since it optimizes for quality, not speed, and starting from chat does not make it faster. The Light tier exists for small, fast tasks.
Conclusion
If your team already runs its work through Claude or ChatGPT, the Poolday MCP server lets you keep that interface for video production. Briefing, follow-ups, questions, file attachments and task checks all work from the chat client, on the same credits as the web app, on every plan. The main gaps are the smaller MCP file limit, the Build-mode default and the lack of a public self-serve setup page.
The practical next step is to pick one small, real job and run it through MCP end to end, then judge whether the briefing loop above covers your workflow. The agent guide explains how jobs and brand kits work, and the blog post on video editing agents explains the category if you are still evaluating tools.