The CoreReflex MCP server exposes the entire creative platform to AI agents through the Model Context Protocol, so a client like Claude can run JSON recipes, queue batch jobs, and pull finished cuts without a human clicking through the app. Instead of a person operating the studio, an agent operates it, planning films, generating assets, and fetching graded results over a single secured seam, this is agentic creative: the studio becomes a tool your AI can call.
What the CoreReflex MCP server is
The Model Context Protocol is an open standard for connecting AI clients to external tools and data. An MCP server publishes a set of capabilities; an MCP client. Claude, or any compatible agent, discovers those capabilities and calls them as tools during a conversation or an automated run.
The CoreReflex MCP server sits on top of the platform's public API. Everything the API can do, the MCP seam can surface to an agent as a callable tool. That means an agent does not need to know how to make HTTP calls or parse responses by hand; it sees clean, named tools, plan a film, run a recipe, check a job, fetch an asset, and uses them the way it uses any other tool.
This fits the larger creative automation story: the same platform a person drives through the editor becomes programmable infrastructure an agent can drive on its own.
What an agent can do over the MCP seam
Because the server is backed by the full platform API, the surface is broad. An agent can:
- Run JSON recipes against the programmable engine that sits over editor state, the same structured workflows that power Autopilot. Our explainer on the programmable JSON engine covers what those recipes can express.
- Trigger the agentic Director loop. PLAN, PRODUCE, CRITIQUE, ASSEMBLE, so an agent can board shots, generate them on Vertex AI, let the per-shot quality gate score and regenerate failures, and assemble a finished cut.
- Queue batch jobs for high-volume generation and poll their status, rather than blocking on a single long-running request.
- Fetch finished assets, graded cuts, images, audio, once a render completes through the deterministic render path.
- Read provenance traces so the agent can audit exactly how each asset was produced: model, prompt, parameters, and score.
The key design choice is that browser surfaces never block on long jobs; the agent plans, kicks off generation, polls, and assembles, the same client-side flow the app itself uses.
How to connect Claude (or any MCP client)
- Create an API key. In your CoreReflex account, mint an API key scoped to the access the agent should have. This key authenticates every call the MCP server makes on your behalf.
- Add the MCP server to your client. In Claude's MCP configuration, register the CoreReflex server entry and provide your API key as the credential. Any MCP-compatible client follows the same pattern, point it at the server and supply the key.
- Confirm the tools appear. Once connected, the client discovers the available tools. You should see capabilities for planning, running recipes, queuing jobs, and fetching assets.
- Issue a natural-language instruction. Ask the agent to plan and produce something, for example, "board a 30-second product teaser and render it." The agent maps that intent onto the MCP tools, runs the Director loop, and returns the finished result.
- Iterate. Because planning, scoring, and editing are free and only generation costs credits, the agent can refine a plan repeatedly before committing to a render.
Full setup details and the current tool list live in the developer documentation.
Securing the connection with API keys
The MCP server is secured by API keys. There is no anonymous access. Each key authenticates and authorizes the agent's calls, so you control exactly what a connected client can do. Treat a key like a credential: scope it to what the agent needs, store it securely, and rotate it if it is ever exposed.
If you are wiring up multiple agents or services, our guide to API keys and auth for automated creative covers how to manage keys, scope access, and keep automated pipelines from stepping on each other.
Build an autonomous video agent
The payoff of the MCP seam is autonomy. Once an agent can plan, generate, gate, and fetch over MCP, you can hand it a goal instead of a click-path. "Produce a weekly product highlight reel from this changelog" becomes a task an agent runs end to end, drafting the brief, boarding the shots, generating on Vertex AI, letting the quality gate catch soft or off-brand shots and regenerate them, and delivering the finished cut.
For a worked example of that pattern, see how to build an autonomous video agent on MCP. The same loop scales from a single ad-hoc request to a fleet of agents producing creative continuously.
Frequently asked questions
How do I connect an MCP client to CoreReflex?
Create an API key in your account, register the CoreReflex MCP server in your client's configuration, and supply the key as the credential. Once connected, the client discovers the available tools, planning, running recipes, queuing jobs, fetching assets, and can act on natural-language instructions. The developer documentation has the exact configuration steps.
What can an agent do over the MCP seam?
Anything the platform API exposes: run JSON recipes, trigger the agentic Director loop, queue and poll batch jobs, fetch finished cuts and other assets, and read provenance traces. In practice that means an agent can take a creative goal and carry it through to a graded, rendered result without a human operating the app.
Is the MCP server secured by API keys?
Yes. Every call is authenticated with an API key. There is no anonymous access. You scope each key to the access an agent needs and can rotate it at any time, so you stay in control of what a connected client is allowed to do.
Do I need to write code to use it?
No. The MCP seam is what lets an agent like Claude operate the studio in natural language, the agent translates your instructions into tool calls. Developers who want tighter control can also call the underlying public API and SDK directly.
Hand the studio to your agent
The CoreReflex MCP server turns a full film studio into a tool your AI can call, plan, generate, gate, and fetch, all over one secured seam. Mint a key, connect your client, and let an agent run the creative work while the quality gate and deterministic render path guarantee the output. Start free with no credit card and connect your first agent today.