Connect the agent you already use
Open the official connector setup page at mcp.character.app and choose your agent or MCP client. It lists the current steps for compatible clients, including Claude Code, Codex, Claude Desktop, ChatGPT and others. Follow that client's instructions and sign in to the Character App account whose work you want to use. The connector is for your account, not access to other members' work.
The connection uses OAuth and the MCP endpoint shown on that page. Client setup changes over time, so the setup page remains the reference for commands and supported sign-in paths. If you use multiple Character App accounts, check which workspace appears before asking the agent to change anything.
Try thisChoose a connection suited to your task
The setup page offers tool presets that narrow the tools exposed to your connection. If you want to inspect a World or understand what you already have, start with the read-only preset. If you want the agent to help make Stories and Clips, choose an appropriate creative preset and review the capabilities returned for that connection.
A preset name is not a promise that every model or generation is available to your account. Permissions, plan and current capabilities still apply. Keep the task specific: explain an existing cast, plan one new scene, or develop one original character. Begin with a small result you can review before asking for an entire series.
Begin by reading your creative workspace
Ask the agent: “Read my Character App creative workspace. Tell me what Worlds, Characters and media I already have, and what this connection can do. Do not change anything yet.” The first tool is get_creative_workspace. The official setup page identifies it as read-only orientation, and the tool returns the current account and capability context.
Use that view to choose an existing World or decide to make something original. The read step does not authorize a generation or a publication. If the workspace is unfamiliar, stop and correct the account or project choice. That small check prevents building a new idea around the wrong cast or starting over with assets you already saved.
Give the agent a concrete creative brief
For an original character, describe a silhouette, personality, outfit and one action. For example: “Plan an original deadpan station attendant in a bright green uniform who treats a lost mitten like a national emergency. Give me three visual directions and a short first scene. Show me the plan before making media.” This gives the agent decisions to explore while preserving your chance to choose.
For an existing World, name the World and Characters and explain what happens next. Ask the agent to use the saved context rather than inventing a replacement cast. Planning can lead to writes or priced work depending on the requested tool; your initial brief should clearly say which work you want done and when to return for review.
Approve an exact quote before paid generation
When the next Character App action is paid, the agent requests the server's quote, shows you the exact credit amount and waits for your explicit approval before confirming it. A creative brief is not permission to spend an unspecified amount. Check the model, operation and price in the returned quote, and ask for a smaller scope if it does not suit you.
If the price changes, the new quote needs a new decision. If a request is still processing or its status is uncertain, the agent should continue checking that existing job rather than submitting another generation. Current account permissions and consent still apply. Compare plans on the pricing page; this guide does not set a fixed price for a multi-step project.
Watch the output and choose what happens next
An accepted request and a completed render are different things. Ask the agent to return the result it can actually inspect, together with its status. Watch the Clip and check the Character, action, framing and sound before choosing the next step. A scene that needs a different expression or action is a creative revision, and a new paid generation still needs its own quote.
Decide separately whether the result should remain private, be exported or be shared. Publishing and public watch links need explicit authorization for the content being shared. You can then ask the agent to continue the same World, plan another scene or help compare the new result with your earlier one. The connector carries the work between sessions; your review decides where it goes.
Common mistakes
Treating a successful connector sign-in as proof that a generation finished.
Giving the agent permission to spend without seeing the exact returned quote.
Assuming one tool preset or account exposes every model and workflow.
Starting a new generation when the existing job is still processing or its response is uncertain.
Publishing a result simply because rendering completed; sharing is a separate decision.
Questions
What is the Character App MCP connector?
It connects a compatible AI agent to your Character App account so the agent can use available tools for your creative workspace, Characters, Worlds, Stories and media. Character App retains the account, permissions, consent and generation process.
Where do I connect?
Start at https://mcp.character.app. It lists the current per-client instructions and the https://mcp.character.app/mcp endpoint. Sign in with the Character App account you want the agent to use.
Does connecting automatically spend credits?
Connecting and the read-only workspace orientation are separate from paid generation. Every paid Character App step needs the exact server quote and your explicit approval before confirmation.
Can I use the Characters and Worlds I already made?
Yes, within your account permissions and the tools exposed by the connection. Begin by reading the creative workspace, choose the saved World or Character, and give the agent the next scene you want to develop.
Does the agent publish everything it makes?
No. A finished render does not itself authorize publication. Ask to review the output, then decide separately whether to export or share it; public audience actions require explicit authorization.
