Frame.io Agent
Enables coding agents to read and write to Frame.io (projects, assets, reviews, comments, and more) with confirmation-gated mutations.
README
Frame.io Agent
by VAXA Studio
Run Frame.io from your coding agent.
Your agent can find your latest cut, summarize the director's notes with timecodes, pull a YouTube reference straight into a project folder, build a contact sheet from your last shoot, and send a curated review link to a client — all without opening Frame.io.
10-minute setup. Works with Claude Code, Cursor, OpenAI Codex CLI, Gemini CLI, or any agent that runs local commands. MCP optional. Read-only by default; every mutation (upload, share, comment) is confirmation-gated.
Unofficial community tool. Not affiliated with, endorsed by, or supported by Adobe or Frame.io. "Frame.io" is a trademark of Adobe.
Quick start (with an agent)
git clone https://github.com/sheldondirector/frameio-agent
cd frameio-agent
Then paste this to your coding agent:
Connect this repo to my Frame.io account, read-only.
1. Read AGENTS.md.
2. Install deps.
3. Set up .env.
4. Run `frameio-agent auth login`.
5. Run `frameio-agent verify`.
6. List my projects.
7. Find my latest cut.
8. Summarize its review comments with timecodes.
Do not print secrets. Prefer --json when parsing.
Quick start (manual)
python scripts/setup.py # install deps
cp .env.example .env # create config
python -m frameio_agent.cli auth login # guided OAuth wizard
python -m frameio_agent.cli verify
python -m frameio_agent.cli projects --json
python -m frameio_agent.cli latest --project <id> --json
python -m frameio_agent.cli comments <file_id> --json
What it does
Read (no confirmation needed):
auth login— guided Adobe IMS OAuth wizard with clipboard auto-detect. First time only.auth status/verify— confirms the connection without printing tokens.projects --json— lists accounts / workspaces / projects.latest --project <id>— newest-updated video assets, recency-first.search "<query>"— account-wide search across all your projects and folders.comments <file_id> --json— normalized review notes (timecode + timestamp_seconds + author + thread).brief --project <id>— one-paragraph project status.
Write (every one is confirmation-gated):
share create <file_id> [<file_id>...] --name "..."— bundle assets into a Frame.io review share. Multi-asset, optional--reviewers email@a.com,email@b.com. Default visibility public;--restrictedfor signed-in only. Prompts y/N before sending;--yesto skip.refs add <url-or-path> --folder <folder_id>— pull a YouTube/X URL via yt-dlp, or upload a local file, or trigger Frame.io's remote_upload for direct URLs Frame.io can fetch. One-stop reference ingestion.contact-sheet --project <id> --out sheet.png— composite thumbnails from a project or folder into a single image grid.
Read-only by default. Mutations only fire after explicit y/N confirmation (or --yes when the user has authorized the action in the current conversation).
Security
- Read-only by default; every mutation is opt-in and confirmation-gated. Share creation, reference uploads, comment posts — all require y/N (or an explicit
--yesthe user authorized). No silent writes ever. - Delete / rename / move / permission-change are still NOT in scope. Even the agent can't accidentally trash your project hierarchy. Those endpoints exist in V4; the CLI just doesn't call them.
- Secrets stay local.
.envand~/.frameio-agent/tokens.jsonare git-ignored. The CLI redacts token-like values from any output. Share passwords are never echoed. - Bring your own credentials. The wizard helps you create an Adobe Developer Console OAuth Web App; nothing is ever sent to the maintainers.
Optional: MCP
If your agent supports MCP, run frameio-agent mcp to expose the same operations as MCP tools. The CLI is the spine — MCP is a thin wrapper.
Troubleshooting
| If you see | Run |
|---|---|
Missing FRAMEIO_CLIENT_ID |
frameio-agent auth login |
Frame.io is not authenticated |
frameio-agent auth login |
Your Frame.io session expired |
frameio-agent auth login |
Found multiple accounts |
Re-run with --account <id> |
Traversal cap reached |
Use latest instead, or pass --folder <id> |
License
MIT. See LICENSE.
Built by VAXA Studio — we make tools for creative teams using coding agents.
Unofficial community tool. Not affiliated with, endorsed by, or supported by Adobe or Frame.io.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。