What is the Valmera MCP server?

It is a Model Context Protocol server that gives an AI model a complete video editing engine. You connect it to an MCP client — the Claude app or Claude Code — and then edit real footage by talking: upload a video, cut the silences, add captions, mix music, reframe it, render a preview, and download the finished MP4, all from the conversation. The endpoint is https://entrepreneur-bot-backend.onrender.com/mcp.

How do I connect Valmera to Claude?

In the Claude app, open Settings → Connectors → Add custom connector and paste https://entrepreneur-bot-backend.onrender.com/mcp. Claude discovers the authorization server, registers itself, and opens a Valmera sign-in and consent page. Sign in with your Valmera email and password, press Allow, and you are connected — there is no token to copy anywhere. Ask it to "list my Valmera projects" to confirm.

Can Claude actually edit video?

On its own, no — a language model cannot open your footage or render a file. Connected to Valmera over MCP it can, because the editing engine does the work: Claude chooses and sequences the tool calls, and Valmera analyzes the footage, applies the edits, renders and exports. What comes back is a real MP4, not a description of one.

How many tools does it expose?

108 at the time of writing — 97 editing tools plus 11 session tools (list and open projects, upload, check indexing, render, export, download, wait on a job). The editing tools are not re-declared for MCP: the engine publishes one registry and the server serves it verbatim, so the model in your session gets exactly what Valmera's own agent has — same names, same schemas, same refusals. A tool whose backing service is unconfigured is hidden from both rather than failing when called, so the live count moves as capabilities are switched on.

How does authentication work?

OAuth 2.1 with RFC 9728 protected-resource discovery, RFC 8414 authorization-server metadata, RFC 7591 dynamic client registration, and mandatory PKCE (S256). Refresh tokens rotate, and replaying an authorization code revokes the whole grant. Claude Code can alternatively carry a static bearer token, which is minted once and stored only as a hash. The project you have open lives on the grant, so a reconnect resumes the same edit.

Does it work with Claude Code, Cursor or other MCP clients?

It is a standard streamable-HTTP MCP server, so any compliant client can talk to it. Claude Code and the Claude app are the two paths we have verified end to end — Claude Code with `claude mcp add --transport http`, the Claude app through the OAuth connector flow. Other clients should work; we have not tested each one, and we would rather say so than list logos.

What happens to slow operations like rendering?

Rendering a preview, exporting a final, or erasing burned-in text can take longer than a tool call should block for. Those return a job number and a wait_for_job tool rather than pretending to be done — your model calls it back and reports honestly in the meantime. Nothing returns a fabricated success.

Can I edit the same project in the studio and from Claude at once?

No, and it is refused in both directions rather than allowed to corrupt the edit: an MCP call is blocked while an agent turn is running on that project, and the studio refuses to start a turn while an MCP tool call is in flight. Every MCP call writes an activity row, so the studio shows the outside model editing live.

How do I upload a large video over MCP?

MCP arguments are JSON, so raw bytes never cross the protocol. Files up to about 64MB can go through the upload tools directly; anything larger uses a small helper script that does a presigned multi-part upload and then hands the finished object to the connector. You can also point Valmera at a link and have it fetch the media itself.

Is my footage modified?

Never. The agent edits an edit decision list, not pixels — your original upload is untouched, every operation creates a new version, and anything cut can be restored. Final exports are rendered from the original file at source quality; previews use a faster proxy.

What do I need to use it?

A Valmera account and an MCP client. If your account was created with Google sign-in it has no password yet, and the connector's consent screen needs one — set it with "Forgot password" on valmera.io first.

Video Editing MCP Server

Valmera is a video editor built as an agent — so the agent can be swapped for yours. Connect the Valmera MCP server to Claude and edit real footage from the conversation you are already in: upload a video, cut the silences, add captions, mix music under your voice, reframe it for Reels, render it, and download the file.

Your model does the thinking. Valmera does the editing — the same engine, the same tools and the same refusals that Valmera's own agent uses.

MCP ENDPOINT

https://entrepreneur-bot-backend.onrender.com/mcp

Streamable HTTP transport · OAuth 2.1 with dynamic client registration and PKCE · 108 tools

How to Connect Valmera to Claude

1
Add the connector
In the Claude app, open Settings → Connectors → Add custom connector and paste the URL https://entrepreneur-bot-backend.onrender.com/mcp — nothing else to fill in.
2
Sign in and allow
Claude discovers the authorization server and opens a Valmera sign-in page. Sign in with your Valmera email and password and press Allow. There is no token to copy. (Google-only accounts should set a password with "Forgot password" on valmera.io first.)
3
Edit by asking
Say "list my Valmera projects" to confirm the connection, then upload footage and describe the edit — "cut the dead air, add karaoke captions, put chill music under my voice, and export it".

Setup is one paste and one sign-in. Indexing a freshly uploaded video takes a while and reports progress; every edit after that reuses the analysis.

Connecting Claude Code

Claude Code can carry a bearer token instead of running the browser flow. Mint one from your Valmera session, then add the server:

claude mcp add --transport http valmera \
  https://entrepreneur-bot-backend.onrender.com/mcp \
  --header "Authorization: Bearer vlm_mcp_..."

The token is shown once and stored only as a hash. The OAuth flow above works in Claude Code too if you would rather not handle one.

What Your Model Can Do With It

Cut

Silence and filler-word removal in one call, arbitrary range cuts and restores, repeated-take detection, and cuts that snap to the musical beat. Everything resolves against a word-level transcript, so "cut the part where I fumble the pricing line" lands on real boundaries.

Caption

Word-accurate burned captions with 11 presets, 12 bundled fonts, karaoke word-pop, per-word emphasis and entrance animations — plus designed text templates, title cards and full-frame colour screens.

Sound

A royalty-free music library or your own track, music that ducks under speech automatically, voiceover, per-item gain, sound effects generated from a description or placed from a file you upload, and loudness mastering to the streaming target.

Picture

Zooms and travelling zooms, automatic punch-ins on vocally stressed words, speed spans from 0.25x to 4x, colour grade presets and continuous grading, finishing effects, transitions at scene changes, overlays and picture-in-picture.

Reframe

16:9, 9:16, 1:1 or 4:5 by crop, pad or blurred pad, with a subject-aware crop that asks a vision model where the subject actually is — plus smooth mid-video aspect shifts.

Repair

Measure and truly erase burned-in captions, watermarks or objects by repainting the pixels and reconstructing the background — not a blur over the top. Deliberate blur, mosaic and black-bar censoring are separate tools when you want the censor visible.

See

Ask a vision model about specific frames of the footage or of any uploaded clip. This is how the agent verifies its own work rather than reporting what it intended.

Render

Fast preview renders while you iterate, then a final export produced from the original upload at source quality, with a download URL back in the conversation.

The complete list — every tool name, what it does and what it refuses — is on the MCP tool reference.

How It Is Built

One registry, not two. The editing tools are not re-declared for MCP. The engine publishes its live registry and the server serves it verbatim, so there is no second list that can drift out of date. A test fails the build if a copy ever appears.
Slow work returns a ticket, not a lie. Renders, exports and pixel repainting exceed a reasonable request timeout. Those answer with a job number and a wait_for_job tool instead of fabricating a completion.
Two editors are refused both ways. An MCP call is blocked while an agent turn is running on the project, and the studio refuses to start a turn while an MCP call is in flight — so the same edit decision list is never written by two editors at once.
Honest-off capabilities. A tool whose backing service is not configured is hidden from the registry rather than exposed and failing on call. Your model is never offered something it cannot do.
Non-destructive by construction. Tools mutate an edit decision list; the renderer touches pixels. Your original upload is never modified, every operation is a new version, and cuts can be restored.
The authorization server refuses carefully. An unregistered redirect URI dead-ends on our own page rather than redirecting, a replayed authorization code revokes the entire grant, and a valid password on an account without access says exactly that instead of "wrong password".

Frequently Asked Questions

Connect It and Edit Something

Create a free Valmera account, paste the endpoint into your MCP client, and edit your first video from the chat.

Create a free account →

See pricing →