Plug ABCD into your AI stack

What you get

• 49 tools — the full ABCD pipeline, one-shot ad/film directors, media primitives, edit/enhance, audio and series planning. Full catalog below.
• Managed rate limits — per-hour caps per API key, live-tunable by ABCD staff.
• Per-user credit charging — pass an ABCD JWT via abcd_token and the generation runs on that user's credits.
• Streamable HTTP transport — the same wire format Claude Desktop and ChatGPT Connectors already speak.

Quickstart

1. Get a URL or a key. Signed-in users: Settings → Connect via MCP shows a personal URL that carries a rotating mcp_sess_… session token — the raw key never travels over the wire. Developers: contact hi@abcdstudio.ai or open the SuperAdmin → API Keys panel and mint a scope-limited mcp_live_… key.
2. Point your MCP client at /api/mcp — either the personal ?token=mcp_sess_… URL, or the bare endpoint with an X-ABCD-API-Key header.
3. Keep the token out of the URL where you can. Cursor, Claude Code (CLI), VS Code, Windsurf and Codex accept headers — send X-ABCD-Token: mcp_sess_… (or Authorization: Bearer mcp_sess_…) to the bare endpoint instead. claude.ai / Claude Desktop connectors, ChatGPT and Kimi take a URL only, so use the masked personal URL and treat it like a password. The Connect page masks the token (👁 reveals it for 20 s) and ↻ Regenerate token invalidates a leaked one instantly without touching your client setup — just paste the new URL.
4. Call any of the tools below. Tools that return a job_id should be polled with poll_job or (for pipeline steps) get_project_status. Every image_url, video_url, audio / stem / export URL and published link comes back as a public https:// link on the ABCD domain (…/api/vibe/media/<file>) that your AI client can open directly — never a localhost address.
5. Watch who connects. ABCD e-mails you and badges Connect the first time your token is used from a new IP or client. Press ✓ This is me on your own tools so they never alert again — even after you regenerate — and anything untrusted that reconnects on a regenerated token is flagged.

{
  "mcpServers": {
    "abcd": {
      "url": "https://abcdstudio.ai/api/mcp",
      "headers": {
        "X-ABCD-API-Key": "mcp_live_YOUR_KEY_HERE"
      }
    }
  }
}

# pip install anthropic openai fastmcp
import asyncio, os
from mcp.client.streamable_http import streamablehttp_client
from mcp import ClientSession

BASE = "https://abcdstudio.ai/api/mcp"
HEADERS = {"X-ABCD-API-Key": os.environ["ABCD_MCP_KEY"]}

async def run():
    async with streamablehttp_client(BASE, headers=HEADERS) as (r, w, _):
        async with ClientSession(r, w) as session:
            await session.initialize()
            # 1. Pitch a film
            res = await session.call_tool("create_film_project", {
                "title": "The Last Signal",
                "genre": "sci-fi",
                "story_concept": ("A radio operator on Mars catches a "
                                   "signal from Earth long after contact "
                                   "was lost."),
                "film_length": "short",
                "aspect_ratio": "16:9",
                "resolution": "1080p",
            })
            print(res.content[0].text)

asyncio.run(run())

# 1. Initialize a session
curl -sN "https://abcdstudio.ai/api/mcp" \
  -H 'X-ABCD-API-Key: mcp_live_YOUR_KEY_HERE' \
  -H 'Content-Type: application/json' \
  -H 'Accept: application/json, text/event-stream' \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize",
       "params":{"protocolVersion":"2025-06-18",
                 "capabilities":{},
                 "clientInfo":{"name":"probe","version":"1"}}}'

# 2. List all tools (save session id from the response header
#    "mcp-session-id" and pass it as Mcp-Session-Id on subsequent calls)
curl -sN "…/api/mcp" \
  -H 'Mcp-Session-Id: SID_FROM_STEP_1' \
  -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/list","params":{}}'

Tool reference

Rate limits

Per-key, per-hour, sliding-window. ABCD staff can override any of these live via db.mcp_rate_limits without a restart. A rejected call returns {ok:false, error:'rate limit reached (…/h)'} — back off with exponential jitter.

Error envelope

Every tool call returns a JSON object whose top-level shape is stable:

{
  "ok": true,
  "session_id": "…", "…": "…",
  "rate_limit": { "used": 3, "cap": 20, "remaining": 17 }
}

{
  "ok": false,
  "error": "backend rejected request (400)",
  "detail": "credit balance too low",
  "error_type": "insufficient_credits",  // when applicable
  "job_id": "…"                          // when a partial run exists
}

Content-flag envelope — when a Seedance render is refused for safety/copyright reasons, error_type is copyright_filter or safety_filter, and a request_id lets support trace the row in db.content_flags.

Video models

Any of these keys work as the model argument on generate_video and animate_scenes. Legacy identifiers auto-alias to the current lineup so old clients keep working:

• seedance-2.0-mini (default · 720p · 24 fps · smart duration · sound on)
• seedance-1.0-pro
• seedance-1.5-pro
• seedance-2.0-fast
• seedance-2.0
• seedance-2.5

End-to-end transcript

The scripted conversation below shows a user directing Claude through a full film project. Each turn shows exactly what tool was called and a compact excerpt of the response. Runnable — swap the story concept, keep the sequence.

{
  "title": "The Last Signal",
  "genre": "sci-fi",
  "story_concept": "A radio operator on Mars catches a signal from Earth long after contact was lost — she has 90 minutes of oxygen to decide whether to reply.",
  "film_length": "short",
  "aspect_ratio": "16:9",
  "resolution": "1080p"
}

{
  "ok": true,
  "session_id": "d94a…c1e2",
  "project_id":  "d94a…c1e2",
  "title": "The Last Signal",
  "status": "planning",
  "pipeline_stage": "script"
}

{
  "session_id": "d94a…c1e2",
  "name": "Commander Riya Vale",
  "description": "South-Asian woman, mid-30s, close-cropped black hair, weather-worn navy Mars-Corps flightsuit with Bengali unit patch, tired but resolute eyes."
}

{
  "ok": true,
  "character_id": "7b3f…9a11",
  "name": "Commander Riya Vale",
  "portrait_url": "data:image/jpeg;base64,/9j/4AA…",
  "description": "South-Asian woman, mid-30s, …",
  "session_id": "d94a…c1e2"
}

{ "session_id": "d94a…c1e2", "wait": true }

{
  "ok": true,
  "script_id": "d94a…c1e2",
  "scene_count": 7,
  "scenes": [
    { "id": "s01…", "order": 1, "title": "Silent radio room",
      "description": "Cold-open on Riya's static-hissing console...",
      "duration": 5 },
    { "id": "s02…", "order": 2, "title": "The signal", "…": "…" },
    …
  ],
  "status": "ready"
}

{ "session_id": "d94a…c1e2", "scene_ids": ["s02…"] }

{
  "ok": true,
  "session_id": "d94a…c1e2",
  "started": [{ "scene_id": "s02…", "order": 2, "status": "queued" }]
}

{ "session_id": "d94a…c1e2" }

{
  "ok": true,
  "session_id": "d94a…c1e2",
  "pipeline_stage": "frames",
  "next_stage": "frames",
  "all_approved": false,
  "characters": [{ "id": "7b3f…", "name": "Commander Riya Vale", … }],
  "scenes": [
    { "id": "s01…", "frame_status": "ready", "approved": false, … },
    { "id": "s02…", "frame_status": "ready", "approved": false, … },
    …
  ],
  "render_progress": { "done": 0, "total": 7, "percent": 0 }
}

{ "session_id": "d94a…c1e2", "scene_id": "s01…", "approved": true }
{ "session_id": "d94a…c1e2", "scene_id": "s02…", "approved": true }
…

{
  "ok": true,
  "scene_id": "s07…",
  "approved": true,
  "session_id": "d94a…c1e2",
  "pipeline_stage": "assembly"
}

{
  "session_id": "d94a…c1e2",
  "model": "seedance-2.5",
  "duration": 6
}

{
  "ok": true,
  "session_id": "d94a…c1e2",
  "model": "2.5",
  "started": [
    { "scene_id": "s01…", "order": 1, "job_id": "vid_a1…" },
    { "scene_id": "s02…", "order": 2, "job_id": "vid_a2…" },
    …
  ],
  "skipped_orders": [],
  "errors": []
}

{
  "session_id": "d94a…c1e2",
  "export_format": "mp4",
  "wait": true
}

{
  "ok": true,
  "session_id": "d94a…c1e2",
  "status": "ready",
  "export_url": "https://abcdstudio.ai/api/vibe/media/d94a_export_1080p.mp4",
  "duration_secs": 42
}