# Frame Lens (npm · framelens)

Eyes and ears for AI agents: measured cuts, easing, palette, beats and transcripts from any video.

- Trust score: 68/100 (medium)
- Change this week: 0
- Registry status: active
- Liveness: live
- Owner verified: no
- Last scored: 2026-11-21

## Components

- npm · `framelens`: 68/100 (this document), [markdown](https://verifymcp.io/servers/makkov-design-framelens/framelens.md), [page](https://verifymcp.io/servers/makkov-design-framelens/framelens)

## Channel facts

- Registry: `npm`
- Package: `framelens`
- Version: `2.1.0`
- Transport: `stdio`

## Trust breakdown

How this component scores in each security and reliability category. Every signal is checked automatically from public evidence about the published package, including repeated runs of it in an isolated sandbox, and we only credit what we can confirm. Scores are 0–100 per category. Scoring method: https://verifymcp.io/docs/scoring (what has changed: https://verifymcp.io/docs/scoring/changelog)

Scored 2026-11-21.

- **Supply Chain Security**: 100/100
  - No malware found by supply-chain analysis.
  - No known CVEs affecting this package version or its production dependencies.
  - No install/post-install scripts declared.
  - No production dependencies, so there is no dependency health to assess.
- **Provenance & Transparency**: 48/100
  - Source repository is publicly reachable at the declared URL.
  - Provenance check failed: no build-provenance attestation is published.
  - Clear OSI-approved license (MIT).
  - Actively maintained (last published 11 days ago).
  - Publishes a security disclosure policy (SECURITY.md).
- **Schema Quality & AI Usability**: 60/100
  - AI-judged instruction clarity (good).
  - Context-footprint check failed: tool/resource definitions use about 3822 tokens (~224/item across 17 items; 17 tools + 0 resources), over budget; trim descriptions and params.
  - Usage-examples check failed: none of the tools include examples.
- **Stability & Change Management**: 0/100
  - Stability not yet verified: not enough scan history yet (needs a 30-day window).
- **Tool Coverage**: 100/100
  - 100% of tools have a non-trivial description (not blank, and not just the tool's name).
  - 100% of tool parameters carry a description.
  - Structured output schemas are declared (12% of tools); any adoption earns full credit.
- **Tool Safety**: 100/100
  - No prompt-injection markers were found in the server instructions, tool names or descriptions we captured.
  - We read all 17 captured tool definition(s), and no name or description among them implies an irreversible operation.
  - An AI judge read all 18 captured unit(s) of tool text and found none that tries to manipulate the model reading it.
- **Capabilities**: 100/100
  - Implements a supported MCP spec version (2025-11-25); the latest is 2026-07-28.

**Unverified: 1 category.** A category scored 0 because we could not verify it: a data source with nothing on this package, evidence we could not reach, or a check we could not run. We only credit what we can confirm.

## Install

### How do I install the Frame Lens MCP server?

Frame Lens runs locally as an npm package, launched with npx -y framelens. Ready-made configuration for Claude, Cursor, VS Code, Codex and 5 more is on this page, copied from each client's own documentation.

### Claude

```bash
claude mcp add makkov-design-framelens -- npx -y framelens
```

### Cursor

```json
{
  "mcpServers": {
    "makkov-design-framelens": {
      "command": "npx",
      "args": [
        "-y",
        "framelens"
      ]
    }
  }
}
```

### VS Code

```json
{
  "servers": {
    "makkov-design-framelens": {
      "command": "npx",
      "args": [
        "-y",
        "framelens"
      ]
    }
  }
}
```

### Codex

```bash
codex mcp add makkov-design-framelens -- npx -y framelens
```

### opencode

```json
{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "makkov-design-framelens": {
      "type": "local",
      "command": [
        "npx",
        "-y",
        "framelens"
      ],
      "enabled": true
    }
  }
}
```

### OpenClaw

```bash
openclaw mcp add makkov-design-framelens --command npx --arg -y --arg framelens
```

### Hermes

```yaml
mcp_servers:
  makkov-design-framelens:
    command: "npx"
    args: ["-y", "framelens"]
```

### Netclaw

```json
{
  "McpServers": {
    "makkov-design-framelens": {
      "Transport": "stdio",
      "Command": "npx",
      "Arguments": [
        "-y",
        "framelens"
      ]
    }
  }
}
```

### Vellum

```bash
assistant mcp add makkov-design-framelens -t stdio -c npx -a -y framelens
```

### Other

```json
{
  "mcpServers": {
    "makkov-design-framelens": {
      "command": "npx",
      "args": [
        "-y",
        "framelens"
      ]
    }
  }
}
```

## Changelog

Every change recorded for this component, newest first. Days that predate change tracking, or that we cannot explain, say so: "we were watching and nothing happened" and "we were not watching" are different claims.

### 2026-11-11 (score 68, 0)

- [functional improvement] Security disclosure: unverified → pass

### 2026-11-10 (score 68)

First indexed and scored.

## MCP tools (17)

### `probe` (~99 tokens)

Open a video

Fetches the video if needed and reports duration, size, frame rate, audio and captions, plus the workspace folder where every later result is written. Call it first; then `plan` for the goal, or `transcript` / `frames` directly.

Input parameters:

- `source` (string, required): The video: a link (YouTube and other sites), a path to a local file, or `inbox` for the newest file in the inbox folder.

### `plan` (~229 tokens)

Plan the analysis for a goal

Classifies the video (talking head, screen, motion graphics, music video, live action) and returns the steps that fit the goal: listen (podcast, lecture), learn (tutorial to guide), find (a moment), style (replicate motion design), sound (sound design), production (how it was made). Pass `goal`, or the user's request in `request` to infer it. Then run the steps in order: each step's `args` are ready to pass (with `source`); a step with `then` says what to add first from an earlier step's result.

Input parameters:

- `goal` (string): The user's goal. Omit to infer it from `request` or the video type.
- `phrase` (string): For `find`: the phrase to look for.
- `request` (string): The user's request in their own words, used to infer the goal.
- `source` (string, required): The video: a link (YouTube and other sites), a path to a local file, or `inbox` for the newest file in the inbox folder.

### `transcript` (~134 tokens)

Word-timed transcript

What is said, with the time of every word when the source has it: captions first, then the configured fallback engine. Writes transcript.json and transcript.md to the workspace and returns the text (long transcripts are shortened; read transcript.md for all of it).

Input parameters:

- `language` (string): Preferred caption language code, e.g. "en" or "ru".
- `max_chars` (integer): Most characters of transcript to return inline.
- `source` (string, required): The video: a link (YouTube and other sites), a path to a local file, or `inbox` for the newest file in the inbox folder.

### `frames` (~420 tokens)

Frames and contact sheets

Extracts frames and arranges them into labelled contact sheets (time, frame number and the words spoken under each picture). Sample with exactly one of: `at` (list of moments), `every` (seconds), `fps`, `every_frame` (every original frame, for motion and transitions), or `around` (a phrase from the transcript; it picks the range and takes `every`, `fps` or `every_frame` with it, 4 fps by default). Work like a loupe: an overview first, then dense frames only around the moments that matter. Sheets are written to the workspace; the first ones are shown inline, open the rest by path.

Input parameters:

- `around` (string): A phrase from the transcript; frames are taken around it.
- `at` (array): Exact moments.
- `columns` (integer): Frames side by side on each contact sheet (1–8, default 4).
- `end`: Range end.
- `every` (number): One frame every N seconds.
- `every_frame` (boolean): Every original frame, with real timestamps (variable frame rate safe).
- `fps` (number): Frames per second over the range.
- `inline` (integer): Sheets to show inline (default 2).
- `occurrence` (integer): Which occurrence of `around`.
- `padding` (number): Seconds before and after `around` (default 1.5).
- `rows` (integer): Rows of frames on each sheet (1–8, default 4); more frames go to further sheets.
- `source` (string, required): The video: a link (YouTube and other sites), a path to a local file, or `inbox` for the newest file in the inbox folder.
- `start`: Range start: seconds (12.5) or M:SS(.mmm).
- `width` (integer): Frame width in pixels (default 320; 240 with 6+ columns).
- `words` (boolean): Put the words spoken at each frame under it.

### `style` (~220 tokens)

Measure the motion style

For replicating a look: finds cuts and gradual transitions (dissolves, fades and dips through black or white, wipes) and the shots between them, measures each shot's palette over its own frames (OKLab, weighted by screen time), reads how much the picture changes on every frame, and lists the moves with their duration in frames and easing family. Writes motion-spec.json (machine-readable, with word and cut anchors), STYLE.md, TIMELINE.md and ANALYSIS.md to the workspace. Then look at the moves with `frames` every_frame, and complete the typography, systems and 'why it works' sections in those files; re-running `style` only refreshes their measured blocks.

Input parameters:

- `end`: Range end (default the end, at most 180 s at a time).
- `source` (string, required): The video: a link (YouTube and other sites), a path to a local file, or `inbox` for the newest file in the inbox folder.
- `start`: Range start (default 0).

### `track` (~325 tokens)

Track an element and fit its exact curve

After `style`, for the moves that matter: follows one element through a window at the full frame rate (subpixel template matching) and fits the exact timing of each property that moves (x, y, scale, opacity): a cubic-bezier, a spring (stiffness, damping, mass) when it overshoots or swings, linear or steps, with the fit error, duration in frames and delay. Window: `move` (a style move number) or `start`/`end`; element: `auto` (the biggest moving region) or a box at time `at`. Writes tracks.json, TRACK.md and an SVG of the curves, and adds the tracked move to motion-spec.json, so `export` uses the exact curve.

Input parameters:

- `at`: Time at which the box is drawn (default the window start).
- `element`: `auto`, or the element's box {x, y, w, h} (top-left and size) in source pixels, or in fractions of the frame when every value is at most 1.
- `end`: Window end (at most 15 s after start).
- `move` (integer): A move from `style` (1 = the first in motion-spec.json); sets the window with some rest around it.
- `source` (string, required): The video: a link (YouTube and other sites), a path to a local file, or `inbox` for the newest file in the inbox folder.
- `start`: Window start (default: just before the style move).

### `audio` (~181 tokens)

Analyze the sound

Sound design and music: loudness (LUFS) over time, silences, onsets (hits, impacts, clicks), tempo and beat grid, and how many cuts land on a beat (when `style` ran first). Returns a spectrogram picture: rising sweeps are risers or whooshes, vertical lines are hits, horizontal bands are tones and music beds. Writes audio-*.json and adds beats to motion-spec.json.

Input parameters:

- `end`: Range end (default the end, at most 10 minutes at a time).
- `source` (string, required): The video: a link (YouTube and other sites), a path to a local file, or `inbox` for the newest file in the inbox folder.
- `start`: Range start: seconds (12.5) or M:SS(.mmm) (default 0).

### `export` (~243 tokens)

Starter project from the reference

Turns the reference's motion-spec.json into a starter project for a renderer: `hyperframes` (index.html + one scene per shot with GSAP eases), `remotion` (Sequences per shot with Easing.bezier), `css` (tokens, keyframes and a preview page), `after-effects` (framelens-reference.jsx: run it in After Effects to build the comp with eased keyframes and cut/beat markers), `lottie` (reference.json with solids, placeholder shapes and bezier easing) or `hypit` (references/<slug>/ANALYSIS.md, TIMELINE.md and transcript.json plus HANDOFF.md for the Hypit video tool, filled from the workspace notes). It carries structure, cut times, colours and move timing; replace the placeholders with the user's own content. Runs `style` first when needed. Afterwards render it and check it with `compare`.

Input parameters:

- `source` (string, required): The video: a link (YouTube and other sites), a path to a local file, or `inbox` for the newest file in the inbox folder.
- `target` (string, required): Renderer to write for.

### `compare` (~96 tokens)

Compare a render with the reference

Checks how close a render is to the reference: duration, cuts matched within 3 frames (missing, extra, late), palette distance, and picture similarity over time with the least similar moments shown side by side. Writes COMPARE.md to the reference's workspace. Fix what it names, render again and compare once more.

Input parameters:

- `reference` (string, required): The reference video.
- `render` (string, required): Path to the rendered video to check.

### `boundaries` (~169 tokens)

Find where the picture changes

Samples the picture and scores how much each sample differs from the previous one. High scores are cut or flash candidates (not editorial labels: check them with `frames` every_frame around each). Also reports how dynamic the video is overall.

Input parameters:

- `end`: Range end (default the end of the video).
- `rate` (number): Samples per second; 10 or more to time cuts precisely.
- `source` (string, required): The video: a link (YouTube and other sites), a path to a local file, or `inbox` for the newest file in the inbox folder.
- `start`: Range start: seconds (12.5) or M:SS(.mmm) (default 0).
- `threshold` (number): Score from which a change is listed (0–1).

### `web` (~244 tokens)

Read a web page's animations

For animations on a live website: opens the page in a headless Chrome or Edge and reads every animation exactly (CSS animations and transitions, Web Animations, Motion, and GSAP when the page exposes it): target, duration, delay, easing curve and keyframes, including ones triggered while scrolling; they are recorded from the first frame, so motion that ends while the page loads counts too. Also reads the real fonts, sizes and colours, takes views down the page, and films the opening motion frame by frame. Writes motion-spec.json and STYLE.md (all values `measured`), so `export` works on it. Needs Node 22+ and Chrome or Edge. For a page behind the user's own login, the user reads it in their browser with the Frame Lens extension ("Read this page's motion"); then pass `inbox` (the newest saved reading) or the path of the .framelens.json it saved, and the same files are written.

Input parameters:

- `url` (string, required): The page address (http or https); or `inbox`, or the path of a .framelens.json page reading saved by the Frame Lens browser extension.

### `exact` (~346 tokens)

Exact animation data (Lottie, Rive, Figma)

Reads the animation parameters straight from the source file instead of measuring pixels: a Lottie file (.json or .lottie, a path or an http(s) link such as lottie.host), a Rive file (.riv, a path or a direct link to the .riv; rive.app pages are not read) or a Figma prototype link with node-id. Reports the composition, every move with its exact cubic-bezier curve, duration and values, the palette, fonts and text, and lists what could not be evaluated (expressions). Writes motion-spec.json and STYLE.md (plus RIVE.md with artboards, animations and state machines for Rive) to the workspace in the same shape as `style`, so the move data can feed the exporters. Figma links need FIGMA_TOKEN (a read-only personal access token). In Claude Code the official Figma MCP tool get_motion_context can also be used: pass its output here as `figma_json` and no token is needed. Pass `export` to also write a starter project (hyperframes, remotion, css, after-effects, lottie, hypit). A video or GIF has no animation data to read: measure it with `plan` or `style` instead.

Input parameters:

- `export` (string): Also write a starter project for this renderer.
- `figma_json`: Figma JSON: a /v1/files/:key/nodes response or get_motion_context output. Use instead of `source`.
- `source` (string): Path or http(s) link to a Lottie .json/.lottie or a Rive .riv file, or a figma.com link with node-id.

### `summarize_youtube_video` (~230 tokens)

Summarize a YouTube video

Summarizes a YouTube video from its frames and audio (sees slides, code, UI and on-screen text, not just captions). Use start/end for one part of the video.

Input parameters:

- `detail` (string): "low" sends frames at low resolution. It saves tokens only on Gemini 2.5 models; Gemini 3 and newer treat it like the default.
- `end`: Analyze up to this point. Same formats as `start`.
- `response_language` (string): Language of the answer, e.g. "Russian". By default the language of the question, otherwise English.
- `start`: Analyze from this point on. Seconds (90), MM:SS (1:30) or H:MM:SS (1:02:03).
- `summary_length` (string): short: 3-5 sentences; medium: overview + timestamped key points; long: detailed sections with visuals.
- `youtube_url` (string, required): YouTube video link (watch, youtu.be, shorts, live or embed) or the 11-character video ID.

### `ask_about_youtube_video` (~235 tokens)

Ask about a YouTube video

Answers a question about a YouTube video using its frames and audio, with timestamps as evidence. Ask specific questions (what is on the slide at 4:10, which settings are clicked, what numbers are shown). Without a question it returns a detailed description.

Input parameters:

- `detail` (string): "low" sends frames at low resolution. It saves tokens only on Gemini 2.5 models; Gemini 3 and newer treat it like the default.
- `end`: Analyze up to this point. Same formats as `start`.
- `question` (string): The question. Omit it to get a detailed description of the video.
- `response_language` (string): Language of the answer, e.g. "Russian". By default the language of the question, otherwise English.
- `start`: Analyze from this point on. Seconds (90), MM:SS (1:30) or H:MM:SS (1:02:03).
- `youtube_url` (string, required): YouTube video link (watch, youtu.be, shorts, live or embed) or the 11-character video ID.

### `extract_key_moments` (~212 tokens)

Extract key moments

Lists the most important moments of a YouTube video with exact timestamps, titles, descriptions and links that open the video at each moment.

Input parameters:

- `detail` (string): "low" sends frames at low resolution. It saves tokens only on Gemini 2.5 models; Gemini 3 and newer treat it like the default.
- `end`: Analyze up to this point. Same formats as `start`.
- `number_of_moments` (integer): How many moments to return (1-50, default 5).
- `response_language` (string): Language of the answer, e.g. "Russian". By default the language of the question, otherwise English.
- `start`: Analyze from this point on. Seconds (90), MM:SS (1:30) or H:MM:SS (1:02:03).
- `youtube_url` (string, required): YouTube video link (watch, youtu.be, shorts, live or embed) or the 11-character video ID.

Output parameters:

- `clip` (object)
- `moments` (array)
- `video_id` (string)
- `video_url` (string)

### `get_video_transcript` (~212 tokens)

Transcribe a YouTube video

Timestamped transcript of what is said in a YouTube video, with speaker labels, plus (optionally) the meaningful text shown on screen. Made by Gemini from the audio and frames, so it works without captions. For videos longer than about 30 minutes, request parts with start/end.

Input parameters:

- `detail` (string): "low" sends frames at low resolution. It saves tokens only on Gemini 2.5 models; Gemini 3 and newer treat it like the default.
- `end`: Analyze up to this point. Same formats as `start`.
- `include_on_screen_text` (boolean): Also list text shown on screen (slides, titles, code).
- `start`: Analyze from this point on. Seconds (90), MM:SS (1:30) or H:MM:SS (1:02:03).
- `youtube_url` (string, required): YouTube video link (watch, youtu.be, shorts, live or embed) or the 11-character video ID.

Output parameters:

- `clip` (object)
- `language` (string)
- `segments` (array)
- `video_id` (string)
- `video_url` (string)

### `list_supported_models` (~32 tokens)

List Gemini models

Lists the Gemini models available to the configured API key and marks the one in use. Useful after a model error.

## Diagnostics

Captured diagnostic sections: Provenance, Dependencies. The full working is on the page: https://verifymcp.io/servers/makkov-design-framelens/framelens#diagnostics

## Score history

- 2026-11-21: 68
- 2026-11-20: 68
- 2026-11-19: 68
- 2026-11-18: 68
- 2026-11-17: 68
- 2026-11-16: 68
- 2026-11-15: 68
- 2026-11-14: 68
- 2026-11-13: 68
- 2026-11-12: 68
- 2026-11-11: 68
- 2026-11-10: 68

## Common questions

### What is the Frame Lens MCP server?

Frame Lens is an MCP server listed in the public MCP registry as io.github.Makkov-design/framelens. Eyes and ears for AI agents: measured cuts, easing, palette, beats and transcripts from any video. This page covers its npm package (framelens).

### Is the Frame Lens MCP server safe to use?

Frame Lens scores 68 out of 100 on VerifyMCP. We found no known CVEs affecting it as of 21 November 2026. It declares no install or post-install scripts. That is a record of what we were able to check automatically, not an endorsement. The category breakdown on this page shows every signal behind the number, including the ones we could not confirm.

### What tools does the Frame Lens MCP server expose?

Frame Lens exposes 17 tools: probe, plan, transcript, frames, style, and 12 more. Their descriptions and schemas cost roughly 3,627 tokens of context every time the server is loaded.

### Is the Frame Lens MCP server still maintained?

Frame Lens is still listed as active in the MCP registry. We last reached this channel on 21 November 2026. Those dates come from our own scans of the registry and the channel itself, not from anything the publisher announced.

### What licence is the Frame Lens MCP server under?

Frame Lens declares the MIT licence, which is OSI-approved. That covers the source only, and says nothing about the cost of any service it calls.

### What did VerifyMCP test on the Frame Lens MCP server?

VerifyMCP ran 15 checks on Frame Lens, covering Malware scan, Known CVEs, Install scripts, Dependency health and 11 more. One item could not be completed, a gap on our side rather than a finding about Frame Lens: Stability. One category scores 0 for that reason, because we only credit what we can confirm. Endpoint Security and Transport & Reachability do not apply to an npm package, so they were not tested. The score does not measure everything, and the scoring documentation lists what it leaves out.

## Links

- npm package: https://www.npmjs.com/package/framelens
- Socket report: https://socket.dev/npm/package/framelens
- Repository: https://github.com/Makkov-design/framelens
- Website: https://makkov-design.github.io/framelens/
- Changelog RSS feed: https://verifymcp.io/servers/makkov-design-framelens/framelens.xml
- Changelog JSON feed: https://verifymcp.io/servers/makkov-design-framelens/framelens/changelog.json
- Trust report (JSON): https://verifymcp.io/servers/makkov-design-framelens/framelens.json
- HTML version of this page: https://verifymcp.io/servers/makkov-design-framelens/framelens
