Add Video Transcription and Clipping to Your AI Workflow — Without Building Any of It Yourself
Wire Weftly into Claude, Cursor, or any MCP-compatible AI agent in minutes. Per-call payment, no API keys, no infrastructure to run.
If you’re already working in Claude, Hermes, OpenClaw, or another MCP-compatible AI tool, you’ve probably hit the same wall: your agent can reason, write, and orchestrate — but the moment a task involves raw audio or video, there’s nothing to hand it to. Building a video processing pipeline from scratch means spinning up infrastructure, managing compute, stitching together transcription APIs, handling file storage, and burning through LLM API credits just to glue it together. For most workflows, that’s not a project you want to take on.
Weftly is an MCP server for video that slots directly into the tools you’re already using. One endpoint. No infrastructure. No API keys to manage. No compute bill at the end of the month.
find_clips tool over MCP — one prompt, ranked clip candidates back.What Weftly Gives Your Agent
Add api.weftly.ai/mcp as a connector in your MCP client and your agent
immediately gains access to five production-grade video tools:
- Transcribe — word-level, timestamped SRT transcripts from audio or video ($0.50 audio / $1.00 video)
- Summarize — AI-generated summary plus the full transcript in a single call ($0.75 audio / $1.25 video)
- Find Clips — ranks the most compelling moments in a long video by impact score, with timestamps and hook text ($2.00)
- Extract Clip — cuts a horizontal segment from a prior job, no re-upload required ($0.50/cut)
- Vertical Clip — cuts a 9:16 short ready for TikTok, Reels, or Shorts from a prior job ($0.50/cut)
Each tool is a single MCP call. Weftly handles the transcription model, the video processing pipeline, file storage, and output delivery. Your agent asks for the result; Weftly returns it.
The Free Claude Code Plugin: One Command to Wire Everything Up
For Claude Code users, Weftly publishes an official free plugin that handles the entire setup — MCP registration, mppx wallet configuration, and payment signing — in a few commands.
/plugin marketplace add woven-record-media/weftly-plugins
/plugin install weftly-setup@weftly
/weftly-setup:weftly-setup --wallet <your-wallet-name>
That’s it. After a Claude Code restart, you have:
weftly:transcribe,weftly:summarize,weftly:find_clips,weftly:extract_clip,weftly:extract_vertical_clip— live from the Weftly MCP servermppx:sign— a local signing proxy that lets Claude handle payment challenges automatically, without any human intervention
The plugin also runs npx mppx skills add to sync mppx’s bundled skills
into Claude’s skill library. Those skills teach Claude the payment dance —
when Weftly returns a payment_required challenge, Claude knows to call
mppx:sign, get the credential, and retry the tool call automatically. New
Weftly tools added to the MCP server are picked up on the next restart with
no plugin update required. The plugin source is MIT-licensed and open on
GitHub.
Payment That Doesn’t Break Your Workflow
This is where MCP tools typically introduce friction: a payment step that interrupts the agent, forces a context switch, or requires manual intervention on every job. Weftly supports the Machine Payments Protocol (MPP) to eliminate exactly that.
When your agent calls a paid Weftly tool, the server returns a payment challenge alongside the job details. If you’ve set up an mppx wallet funded with USDC on the Tempo network, your MCP client signs the challenge and retries automatically — no browser tab, no interruption, no human in the loop. The agent pays, the job runs, and you get the result. Weftly never sees your private key; signing happens entirely on your machine.
Most jobs run $0.50–$2.00. Fund a wallet once with a few dollars and it covers a significant volume of work. On-chain receipts are available for every transaction.
Don’t want to set up a wallet? No problem. Weftly also supports card payments through Stripe Checkout. When you’re working in a browser-based MCP host — Claude Desktop, claude.ai, or VS Code Copilot — the payment challenge comes back as a Stripe Checkout link instead. Click it, pay by card or Apple/Google Pay, and your agent picks up where it left off. No wallet, no USDC, no crypto required.
No Infrastructure. No API Keys. No Compute.
This is the part worth dwelling on. Standing up a video processing stack yourself means:
- A transcription provider and its API key rotation
- A video encoding service or self-managed FFmpeg workers
- Object storage with lifecycle policies
- A job queue and status polling layer
- A payment integration if you want to run it as a service
Weftly is all of that, already built and running. You don’t manage the LLM
that generates summaries. You don’t pay for idle compute between jobs. You
don’t hold client files longer than needed — for MCP / MPP agent jobs,
Weftly auto-deletes uploaded files within 4 hours (72 hours for Find
Clips jobs, to keep the source available for derivative extract_clip and
extract_vertical_clip cuts). Your agent just calls the tool.
Adding Weftly to Your MCP Client
The setup is the same across any MCP-compatible host:
https://api.weftly.ai/mcp
In Claude Desktop or claude.ai, go to Settings → Integrations → Add custom connector and paste the endpoint. In Cursor, OpenClaw, Goose, or your own client, add it as a standard MCP Streamable HTTP server — no session ID, no registration, no API key. Payment-per-call is the auth.
The server exposes a full tools/list response with JSON Schema for every
tool, so your client can discover capabilities at runtime without reading
documentation. The server is stateless — no Mcp-Session-Id is issued or
required.
The browser interface at chat.weftly.ai is available if you want to run a job manually first and see exactly what a Weftly output looks like before wiring it into an agent workflow.
For the broader case for pay-per-job AI tools, see Weftly is the Post-Production Tool Small Creators Have Been Waiting For. For the professional-services angle, see Accurate, Affordable Transcription for Lawyers, Journalists, and Researchers.
MCP endpoint: api.weftly.ai/mcp · Claude Code plugin:
github.com/woven-record-media/weftly-plugins
· OpenAPI spec:
api.weftly.ai/.well-known/openapi.json
· Pay per job · No subscription ·
weftly.ai
Common questions
- What is MCP?
- The Model Context Protocol is an open standard for connecting AI agents to external tools and data. Weftly exposes its transcription, summarization, and clip-cutting capabilities as MCP tools, so any MCP-compatible host (Claude, Claude Desktop, Cursor, VS Code Copilot, etc.) can call them directly.
- Do I need API keys to use Weftly's MCP server?
- No. The Weftly MCP server uses per-call payment via the Machine Payments Protocol — you pay $0.50 to $2.00 per job at the moment of use, either by card through Stripe or with USDC on Tempo. There is no signup, no API key to manage, and no monthly bill.
- How do I install Weftly in Claude Code?
- Run `/plugin marketplace add woven-record-media/weftly-plugins` then `/plugin install weftly-setup@weftly` and `/weftly-setup:weftly-setup --wallet <your-mppx-wallet>`. The plugin wires up MPP signing automatically. The full guide is at weftly.ai/mcp/.
- Can my agent reuse a transcript across multiple clip cuts?
- Yes. After find_clips runs ($2.00), the source video stays cached for 72 hours. Pass the find_clips job_id as source_job_id to extract_clip or extract_vertical_clip and each cut is $0.50 with no re-upload and no re-transcription.
Try Weftly
Transcribe, summarize, and clip videos in minutes. Pay per job, no subscription.
Try it in the browser