Make videos from Claudeand Codex

Connect the MCP server and ask for a shot in plain words. It quotes first, asks before it spends, and returns a link.

Real renders not open yetServer URLhttps://mcp.pershot.app/mcp

Connecting your own AI agent is open to every account and needs no key. Developer API keys, for automating your own work from code, are by application.

Example conversation

You

Turn this product photo into a 9:16 ad, 5 seconds.

AI agent

Here is the plan: Seedance 2.0 Fast, 720p, 9:16, 5 s, first frame from your photo. Quote: 34 credits (≈ $0.34). Go ahead?

quote → { "model": "seedance-2.0-fast", "resolution": "720p", "aspect_ratio": "9:16", "duration": 5 }

You

Yes, render it.

AI agent

Done. Here is the clip: a signed download link and a thumbnail. 34 credits charged; your balance is 1,966 credits.

Connectin a minute

Pick your client. Sign-in uses OAuth, so no key is pasted into a chat.

Claude Code

Add the server, then type /mcp in Claude Code and sign in with your account in the browser.

Terminal

claude mcp add --transport http pershot https://mcp.pershot.app/mcp

claude.ai and Claude Desktop

Open Settings, then Connectors.
Add a custom connector and paste the server URL.
Connect, sign in to your account and choose what it may do.

Codex CLI

Add the server, then sign in with codex mcp login.

Terminal

codex mcp add pershot --url https://mcp.pershot.app/mcp
codex mcp login pershot

Or edit ~/.codex/config.toml

For automation: the same server with a connection code from AI connections. The code stays in the environment variable, never in the file.

config.toml

[mcp_servers.pershot]
url = "https://mcp.pershot.app/mcp"
bearer_token_env_var = "PERSHOT_MCP_CODE"

For scripts and agents that cannot sign in, generate a connection code under AI connections and send it as a Bearer token. It spends your creative balance too.

The toolsit adds

A thin layer over the public API: the same quotes and ledger rules as everywhere. Signed in, it spends your creative balance, never the developer API balance.

Tool	What it does	Spends credits
`list_models`	Available models, capabilities and status	No
`get_pricing`	Public prices and your account's prices	No
`quote`	The plan, line items, maximum charge and expiry	No
`use_preset`	Turns a preset into parameters for quote	No
`create_upload`	An upload URL, or registers a public https link	No
`generate_video / generate_image`	Submits with a quote_id or max_credits; returns a job id	Yes
`get_job / wait_for_job`	Job status; wait_for_job waits up to 60 seconds	No
`cancel_job`	Asks to cancel a job	No
`get_balance`	The balance this connection spends, and its remaining budget	No

Your creditsyour limits

Whose money: It spends your creative balance, from creative top-ups or a membership. We never advance credit.
At zero: It stops when your creative balance runs out (insufficient_credits). It never switches to the developer API balance.
Per-connection cap: You can set a monthly cap per connection, for a teammate or an automation.
Confirmation: Paid tools are marked as such, so your client asks you before running one.
Retries: Paid calls need a quote or a cap and an idempotency key, so an agent's retry never charges twice.
Auto top-up: MCP spending never triggers auto top-up unless you set a cap for that connection.

Safetybuilt in

Scopes: Read-only, generate, and billing read-only permissions.
Links: Registering an external link never reaches private addresses.
Moderation: The same content checks as the web studio.
Traceable: Every job records where it came from (mcp:claude, mcp:codex), for support and your own reports.

MCPquestions

Which clients work?

Claude (claude.ai, Claude Desktop, Claude Code) and Codex, plus any client that speaks MCP over Streamable HTTP.

Is this an integration made with Anthropic or OpenAI?

No. It is a standard MCP server that their clients can connect to. We are not affiliated with either company.

Does the MCP server cost extra?

No. Connecting is free, and each render costs the same as in the studio, from your creative balance.

When can I connect?

When real renders open. Leave your email on a model page and we write once.