Agents and MCP

---
name: eden-gen
description: Generate images, video and audio through Eden Gen (Seedance, Kling, Veo, Nano Banana, GPT Image). Use when the user asks for a shot, still, storyboard frame, video clip or voice line.
---

# Eden Gen

Use the Eden Gen MCP tools (server https://edengen.ai/mcp). Their names start with takes_.

1. takes_list_models, pick an efficient draft model first (seedance-2.0-mini, nano-banana-2.1).
2. takes_list_projects, bill to the project the user names.
3. takes_estimate, tell the user the price. If needs_confirmation, ask before creating.
4. takes_create with a directing prompt: camera, beat, motion, light. Video ends with "no music", one shot adds "no cut".
5. takes_get until done. Failed takes are refunded; report the reason.
6. Keep the take ID in every file name.

import json, os, requests
from openai import OpenAI

TAKES = "https://edengen.ai"
H = {"Authorization": f"Bearer {os.environ['TAKES_API_KEY']}"}
spec = requests.get(f"{TAKES}/v1/tools?format=openai").json()
llm = OpenAI(api_key=os.environ["DEEPSEEK_API_KEY"], base_url="https://api.deepseek.com")

def call(name, args):
    r = spec["routes"][name]
    path = r["path"].replace("{id}", args.pop("id", ""))
    res = requests.request(r["method"], TAKES + path, headers=H,
                           json=args if r["method"] == "POST" else None,
                           params=args if r["method"] == "GET" else None)
    return res.text

msgs = [{"role": "user", "content": "Draft a 5 s shot of a lighthouse in a storm. Show me the price first."}]
while True:
    out = llm.chat.completions.create(model="deepseek-chat", messages=msgs, tools=spec["tools"]).choices[0].message
    msgs.append(out)
    if not out.tool_calls:
        print(out.content); break
    for tc in out.tool_calls:
        msgs.append({"role": "tool", "tool_call_id": tc.id,
                     "content": call(tc.function.name, json.loads(tc.function.arguments))})