Connect Your AI Agent (MCP)

Plug Claude Code, Codex, or any MCP client into Zonic — browse projects, query footage understanding, and run Dekiru edits from the agent you already use.

One Endpoint

Zonic is an MCP server. POST /api/mcp speaks MCP over Streamable HTTP (JSON-RPC 2.0), authenticated with the same x-api-key header as every other API endpoint. It is account-scoped: one connection covers all your projects, with each tool taking a pid argument.

Use /api/mcp to let your agent operate Zonic across projects — including running Dekiru edits.
The per-project /api/projects/{pid}/mcp endpoint remains for read-only retrieval pinned to a single project.

Connect A Client

Any MCP client that supports Streamable HTTP with custom headers works. The two most common:

claude mcp add --transport http zonic https://zonic.tech/api/mcp \
  --header "x-api-key: $API_KEY"

# then, in any session:
#   "list my zonic projects"
#   "which clips show the keynote demo?"
#   "make a 60s recap from my latest project"

Use the API key from your Zonic account. The key authorizes exactly the projects you own — your agent cannot see or edit anyone else's footage.

Tools

Call tools/list to discover the current set. It spans the full workflow:

list_projects — entry point; every other tool takes a pid from here.
Footage retrieval — search_scenes, search_script, search_labels, get_transcript, get_word_transcript, list_people, get_video_descriptions, get_video_metadata, get_scene_understanding, get_key_quotes, list_videos.
get_sequences — list edit sequences, or fetch one full timeline by index.
generate_rough_cut / revise_sequence — run the Dekiru editing agent server-side.
get_job — poll a running edit until it finishes.

Editing Jobs

A Dekiru edit takes minutes, so the editing tools return immediately with ajob_id while the agent works. Poll get_job every 15–30 seconds; on done it returns the agent's answer, the new sequence index, and a link to open the sequence in the Zonic editor.

curl -X POST "https://zonic.tech/api/mcp" \
  -H "x-api-key: $API_KEY" -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0", "id": 1, "method": "tools/call",
    "params": {
      "name": "generate_rough_cut",
      "arguments": { "pid": "YOUR_PID", "brief": "60s energetic event recap, open on the venue" }
    }
  }'

# → { "job_id": "...", "status": "running", ... }  — then poll:

curl -X POST "https://zonic.tech/api/mcp" \
  -H "x-api-key: $API_KEY" -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0", "id": 2, "method": "tools/call",
    "params": { "name": "get_job", "arguments": { "job_id": "JOB_ID" } }
  }'

Editing runs on Zonic's servers and consumes your Zonic credits, exactly like edits made in the app. Your agent client never needs Zonic's editorial models or prompts — the intelligence stays server-side.

Project Modes

Footage QA and sequence reads work for every project mode — cloud, hybrid, and local.
Editing tools require a cloud or hybrid project, where the Dekiru agent runs server-side.
For a local-mode project the editing agent lives on your own machine — use the Zonic app there, or switch the project to hybrid.

Back: Dekiru agent Next: API reference