streamable-httpMITupdated 2mo ago
Flask is the feedback layer for video, built for the agentic loop: your agent uploads a render and shares the link instantly. The reviewer doesn't type - they hit record and talk through the video (voice, camera, screen, drawing on frames), and Flask turns the recording into structured, timestamped comments with transcripts. The agent reads that feedback and iterates, pushing each revision as a new version of the same asset. Typed comments work too - recordings are the advantage, not a requirement.
What can you do with flask?
Flask plugin for Claude Code
Flask is the feedback layer for video, built for the agentic loop: your agent uploads a render and shares the link instantly. The reviewer doesn't type - they hit record and talk through the video (voice, camera, screen, drawing on frames), and Flask turns the recording into structured, timestamped comments with transcripts. The agent reads that feedback and iterates, pushing each revision as a new version of the same asset. Typed comments work too - recordings are the advantage, not a requirement.
This plugin connects the Flask MCP server and teaches the agent the full review loop.
Install
One command connects the Flask MCP server (Claude Code, Cursor) and installs the review-loop skill for any agent that reads skills:
npx flask-feedback
Prefer to wire things yourself? Per-client instructions below.
Claude Code
/plugin marketplace add tryflask/skills
/plugin install flask@flask
Then authenticate once: /mcp -> flask -> complete the browser sign-in.
No plugin manager? Connect the MCP server directly:
claude mcp add --transport http flask https://api.flask.do/api/mcp/mcp
Cursor, Codex, and other agents
Install the review-loop skill (works across skills-compatible clients):
npx skills add tryflask/skills
Then connect the MCP server:
- Cursor:
- claude.ai / Claude Desktop: Settings -> Connectors -> add
https://api.flask.do/api/mcp/mcp - Any MCP client: Streamable HTTP at
https://api.flask.do/api/mcp/mcp(OAuth sign-in opens in the browser on first use)
What the agent can do
| Tool | What it does |
|---|---|
contents, search, recent_activity |
Browse folders/assets, search, latest team feedback |
feedback_list, feedback_get |
Read feedback with tags, timestamps, recording transcripts (transcript: "full" for whole recording) |
wait_for_feedback |
Long-poll - returns new feedback the moment it's left |
get_annotated_frames |
Returns the video frames a recording pointed at / drew on, as images with the drawing rendered in, plus the transcript marked [FRAME N] - resolves "this"/"here"/drawn-circle references to actual pixels |
upload_file_start / upload_file_complete |
Upload a local video (5GB max) via presigned URL; share link available the instant the upload starts |
upload_video |
Import from a public URL or Google Drive link |
version_of (param on uploads) |
Upload as a new version of an existing asset - one stable link for the whole iteration |
asset_status, tags, permission_get |
Processing status, tag distribution, folder access |
The server is read-only except for uploads - it can never edit or delete anything.
The loop in practice
agent renders video -> upload_file_start -> user gets flask.do link instantly
user records feedback on the video -> wait_for_feedback returns it (transcribed)
agent implements changes -> uploads v2 with version_of -> same link shows v2
Docs and support
- Server documentation: https://flask.do/mcp
- Support: hello@flask.do
This repo is kept in sync with the MCP server. Tool list and behavior described here mirror https://flask.do/mcp, which is the source of truth.
Install
Add flask to your client. Pick the one you use.
claude mcp add --transport http flask https://api.flask.do/api/mcp/mcpcodex mcp add flask --url https://api.flask.do/api/mcp/mcp{
"mcpServers": {
"flask": {
"url": "https://api.flask.do/api/mcp/mcp"
}
}
}Add to `~/.cursor/mcp.json`, or `.cursor/mcp.json` for a single project.
{
"servers": {
"flask": {
"type": "http",
"url": "https://api.flask.do/api/mcp/mcp"
}
}
}Add to `.vscode/mcp.json` in your workspace.
{
"mcpServers": {
"flask": {
"url": "https://api.flask.do/api/mcp/mcp"
}
}
}Add to `claude_desktop_config.json`, then restart Claude Desktop.
{
"mcpServers": {
"flask": {
"serverUrl": "https://api.flask.do/api/mcp/mcp"
}
}
}Add to `~/.codeium/windsurf/mcp_config.json`.
Score
39 / 100
Incomplete
- Documentation22/25
- Maintenance16/25
- Trust16/20
- Capability0/15
- Install experience12/15
- Documents what it does and how to connect
- Has a resolvable package or endpoint
- Exposes at least one tool, prompt or resource
- README has substantive content
- Includes a code example
- Documents its configuration
- Mentions credentials or security posture
- Last commit 55 days ago
- Has a release history
- Repository is not archived
- Licensed MIT
- Namespace verified in the official MCP registry
- Claimed by its owner
- Published under an organisation
- 0 tool(s) documented
- Provides prompt templates
- Provides resources
- 6 documented install method(s)
- Published to a package registry
- Offers a hosted endpoint — no local install
Version history
| Versions | Published |
|---|---|
| 1.1.1Latest | Jul 8, 2026 |