Reference

All 49 tools

The complete VidMee tool surface. Over MCP these are tool calls; over REST each is POST /v1/jobs {"operation": "<name or alias>", "input": {...}}. Names are underscore-cased for the tool-use API; the dotted alias also works as a REST operation.

Start with the hero tools (tagged below): edit_image, inspect, report, storyboard_video, sweep_site, test_flow, verify_against_reference. Each fuses many primitives into one verdict, so one call usually replaces five.

QA & flows

inspect

heroqa.inspect

Render a page (or drive an open session) and return ONE fused verdict: UI/UX critique, real axe-core WCAG, runtime errors, visual-memory regression check, and next actions. The hero call.

params: source, goal, viewport, compare_to, save_baseline, client_ref

sweep_site

heroqa.sweep_site

Crawl a whole site and judge every page: per-page findings plus a severity rollup. Baselines live in memory so re-renders diff against last-known-good. Nightly QA in one call.

params: source, max_pages, viewport, client_ref

test_flow

heroqa.test_flow

Drive a scripted multi-step browser flow (click, type, navigate) and return a pass/fail verdict with per-step evidence. Logged-in journeys included.

params: start_url, steps, viewport, login, client_ref

Perceive

analyze_document

specialistperceive.document

Read a PDF or design export into structured data.

params: source, pdf_base64, max_pages, questions, client_ref

analyze_screenshot

specialist

Judge a screenshot you already have (send local files as image_base64). Returns critique + asset_id.

params: image_base64, image_url, questions, tags, client_ref

analyze_video

specialistperceive.video

Judge a video journey frame-by-frame.

params: source, asset_id, frames, questions, client_ref

analyze_visual

specialistperceive.analyze

Analyze a stored asset by id against your questions.

params: asset_id, questions, comparison_asset_id, client_ref

capture_element

specialist

Full-resolution screenshot of a single element by selector (e.g. lift a logo).

params: url, selector, viewport, client_ref

check_crawlability

specialistperceive.crawlability

Measure how much of a page is invisible to no-JS crawlers and AI agents (AEO readiness).

params: source, client_ref

creative_self_qa

specialistperceive.creative_qa

One-call creative gate: aspect ratio + brand fidelity + text/quality in a single verdict.

params: candidate, reference, expected_aspect, criteria, client_ref

extract

specialistperceive.extract

Schema-guided structured extraction: pull typed fields off a rendered page.

params: source, fields, schema, instructions, viewport, client_ref

og_check

specialistperceive.og_check

Validate OpenGraph / Twitter share-preview meta and image.

params: source, client_ref

render_and_analyze

specialistperceive.render

Render a URL/HTML and return a UI/UX + accessibility critique with concrete fixes, deterministic runtime-error findings, and a signed screenshot URL.

params: source, source_type, viewport, questions, wait_for, namespace, client_ref

validate_responsive

specialistperceive.responsive

Flag layout breaks, overflow, and hidden CTAs across viewports.

params: source, source_type, viewports, client_ref

Compare

verify_against_reference

herocompare.verify_reference

Grade a candidate against a real reference (mock, brand, prior screenshot) for fidelity. Catches convincing fakes that aesthetic QA passes.

params: candidate, reference, criteria, client_ref

compare_screenshots

specialistcompare.screenshots

Pixel + semantic diff of two stored assets for regressions, with changed regions.

params: before_asset_id, after_asset_id, criteria, client_ref

Memory

create_baseline

specialist

Pin a stored asset as a named baseline to diff future renders against.

params: name, asset_id, viewport, description

find_similar_visuals

specialistmemory.similar

True image-to-image similarity (DINOv2) for 'more like this' and near-duplicate / regression detection.

params: asset_id, image_base64, namespace, limit

ingest_visual

specialistmemory.ingest

Pull an external image into visual memory with tags and an optional TTL.

params: source_url, context, tags, namespace, ttl_seconds, client_ref

list_baselines

specialist

List the baselines saved for your project.

search_visual_memory

specialistmemory.search

Semantic search over remembered assets.

params: query, tags, asset_type, namespace, limit

Generate & create

edit_image

herogenerate.edit

Edit a REAL photo from a plain instruction while preserving the real product/branding and designing the rest. Built-in fidelity gate + retry. Nano Banana or FLUX Kontext.

params: source, instruction, keep, references, provider, max_attempts, lock_regions, client_ref

storyboard_video

herogenerate.storyboard

Shot list to one finished multi-shot video: generate every shot through the gen-quality path, then stitch. Camera direction + supplied audio.

params: shots, provider, quality_hint, aspect_ratio, audio_url, audio_asset_id, captions, transition, resolution, fps, analyze, client_ref

compose_video

specialistgenerate.video

Assemble finished video from clips/frames (+ motion, transitions, supplied audio, burned captions).

params: clips, audio_url, audio_asset_id, transition, transition_duration, default_motion, captions, resolution, fps, fit, analyze, client_ref

composite_image

specialistgenerate.composite

Deterministic overlay on a real photo: text, logo, or clone-remove a region. Swap branding pixel-exactly.

params: base, elements, description, client_ref

generate_until_pass

specialistgenerate.until_pass

Regenerate until it passes a quality/fidelity bar; escalates across providers and returns the best attempt even if none fully pass.

params: prompt, criteria, max_attempts, media_type, budget_hint, reference_asset_id, client_ref

generate_visual

specialistgenerate.visual

Text to image/video via the best provider (Zark/fal/Replicate), with task-aware model + tuned params. Pass provider to force one, quality_hint for cinematic.

params: prompt, media_type, style_reference_asset_id, provider, quality_hint, budget_hint, namespace, public, size, brand_kit, template, title, client_ref

list_models

specialist

The model catalog: per media, per task, which model each provider uses + tuned params.

list_providers

specialist

Discover the generation providers, what each is best for, and how to override.

render_template

specialistgenerate.template

Deterministic on-brand statics (cards, OG images) from a template + tokens.

params: template, tokens, content, description, namespace, client_ref

Report & tickets

report

heroreport.build

Turn an inspect/test_flow result (or a source URL) into a structured markdown QA report.

params: result, source, title, client_ref

create_ticket

specialist

File a finding straight into GitHub / Linear / Slack from a QA result.

params: target, title, result, body, repo, token, webhook_url, labels

Browser (low-level)

browser_act

low-level

Act in the session: click / type / select by ref or selector.

params: session_id, action, ref, selector, text

browser_close

low-level

Close the browser session and free the resource.

params: session_id

browser_navigate

low-level

Navigate the open session to a URL.

params: session_id, url

browser_observe

low-level

Return a compact semantic read of the current page (not raw DOM) with actionable refs.

params: session_id

browser_open

low-level

Open a persistent, stateful browser session the agent drives across steps (optionally restoring saved auth).

params: viewport, start_url, storage_state, login

browser_save_session

low-level

Save the session's auth/storage state for later reuse (logged-in flows).

params: session_id

browser_screenshot

low-level

Screenshot the current session state.

params: session_id

Account & results

delete_login

specialist

Delete a vaulted login by name. Cost: free (utility/read).

params: name

estimate

specialist

Price a job BEFORE running it — no job created, no charge (#4). Generation is pass-through provider cost + margin (an estimate that settles on the real cost); analysis is a flat credit price. Pass the same `operation` + `input` you'd send t

params: operation, input

get_brand_kit

specialist

Read one stored brand kit (palette, wordmark, style, logo asset) by slug. Cost: free (utility/read).

params: slug

get_result

specialist

Fetch the result of a long-running job by id.

params: job_id

get_wallet

specialist

Check analysis-credit and generation-cash balances before spending.

list_brand_kits

specialist

List this project's stored brand kits (slug + name + palette + style). Pass a slug as `brand_kit` to generate_visual (with a `template` + `title`) to produce in that house style — no hand-carried brand prompt. Create/extract kits via the RE

list_logins

specialist

List your vaulted logged-in sessions (names + domains + expiry — never the stored credentials). Pass a name as login="<name>" to browser_open / test_flow / inspect. Cost: free (utility/read).

record_outcome

specialist

Tell VidMee whether a prior judgment was right (confirmed_real / fixed / false_positive). How accuracy is proven.

params: asset_id, outcome, analysis_id, source, external_system, external_ref, detail, occurred_at, idempotency_key

save_login

specialist

Vault a LOGGED-IN browser session under `name` so you can resume authenticated QA/browsing by name later — log in ONCE, reuse forever (no re-holding cookies). Three ways to provide it: `from_session_id` (capture a live browser_open session

params: name, from_session_id, cookies, storage_state, expires_days, label