webEmbedding
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| WEB_EMBEDDING_TELEMETRY | No | Set to '1' to enable telemetry | |
| WEB_EMBEDDING_NO_TELEMETRY | No | Set to '1' to disable telemetry even if prompted | |
| WEB_EMBEDDING_TELEMETRY_LOG | No | File path to log telemetry events locally | |
| WEB_EMBEDDING_TELEMETRY_PROMPT | No | Set to '0' to skip telemetry prompt during install | |
| WEB_EMBEDDING_TELEMETRY_ENDPOINT | No | URL to send telemetry events to |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
webEmbedding clones a web page from its URL: it captures the live page in a real browser, reuses the original (iframe/embed/source) when allowed, otherwise writes a self-contained offline HTML snapshot or rebuild, and verifies the result against screenshots at desktop, tablet, and mobile breakpoints.
Use these tools when the user gives a URL and wants the page cloned, copied, recreated, mirrored, archived as a faithful offline copy, embedded, or checked for visual fidelity (e.g. "clone this site", "make it look exactly like", "그대로 가져와줘", "완전 똑같이").
Tool choice:
clone_reference_url: default for "clone/copy this page" — runs inspect, capture, rebuild, and self-verify in one call and returns the clone path plus scores.
audit_reference_url: cheap preflight (no browser) to check whether reuse, local capture, login, or manual review is needed.
inspect_url / discover_embed_candidates: frame policy and reusable embed/source URLs.
generate_embed_snippet: iframe snippet once a reusable URL is known.
capture_reference_bundle + verify_fidelity_report: capture evidence and compare a reference against your own implementation.
Do not use for scraping/summarizing content, original designs without a reference URL, or bypassing logins, paywalls, captchas, or licenses.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| detect_runtime_capabilitiesA | Report whether node, Playwright, and a usable Chrome/Chromium binary are available for session-aware capture. |
| inspect_urlA | Inspect a URL before cloning: fetch it, read frame policy (X-Frame-Options/CSP), metadata, and platform signals, and summarize whether it can be embedded, sourced, or must be rebuilt. |
| discover_embed_candidatesB | Extract likely embed, preview, viewer, remix, and export candidates from a page. |
| trace_runtime_sourcesB | Use a browser runtime to trace preview, embed, and scene URLs for a single viewport, with optional session-aware capture when Playwright is available. |
| classify_clone_modeB | Decide whether a reference should be embedded, sourced, rebuilt, or blocked. |
| audit_reference_urlA | Preflight check before cloning or embedding a URL, without launching a browser. Returns whether the page can be reused/embedded as-is, needs local browser capture, needs a logged-in session, needs manual review, or should be blocked, plus the next tool to call. |
| generate_embed_snippetA | Generate a ready-to-paste iframe embed snippet (HTML or Next.js) for a URL that is known to allow framing. |
| capture_reference_bundleA | Capture evidence for a URL with a real browser: runtime HTML, full-page screenshots, DOM and computed-style summaries, assets, network/HAR metadata, interactions, and optional desktop/tablet/mobile breakpoints. With output_dir and capture_html it also writes an offline HTML snapshot. Use when you will build or verify a page yourself. |
| plan_reproduction_pathC | Produce a source-first execution plan using the current policy and capture bundle state. |
| verify_fidelity_reportA | Compare a reference capture against a candidate (e.g. your own implementation) and score visual, DOM, computed-style, and interaction similarity, listing the biggest differences to fix. |
| build_reproduction_bundleA | Turn a capture bundle into an exact-reuse or reproduction bundle with persisted embed files when a trusted reuse path exists. |
| build_rebuild_scaffoldB | Generate a bounded rebuild scaffold from an existing capture bundle when exact reuse is unavailable. |
| clone_reference_urlA | Clone a web page from its URL in one call. Use when the user asks to clone, copy, recreate, mirror, or make a page look exactly like a URL. Captures the live page in a real browser, reuses the original when it can be embedded, otherwise writes a self-contained offline HTML snapshot (inlined CSS, shadow DOM, localized fonts/images) or a rebuild, and self-verifies it against desktop/tablet/mobile screenshots. Returns the clone entrypoint, output directory, and fidelity scores. |
| enqueue_clone_jobA | Persist a source-first clone job into a filesystem-backed async queue for later worker execution. |
| list_clone_jobsB | List persisted clone jobs from the filesystem-backed async queue. |
| load_clone_jobB | Load one persisted clone job by id. |
| cancel_clone_jobA | Move a queued or retry-wait clone job to cancelled. |
| run_clone_jobA | Run one clone queue job, or the next due queued/retry-wait job when job_id is omitted. |
| replay_har_requestsB | Replay request specs against a standard HAR, near-HAR, or webEmbedding network manifest and return deterministic response matches. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 19 tools
Several tools overlap around preflight decision-making: inspect_url, audit_reference_url, and classify_clone_mode all assess whether a URL can be embedded, sourced, rebuilt, or blocked. The granular pipeline tools (capture_reference_bundle, plan_reproduction_path, build_reproduction_bundle, build_rebuild_scaffold) versus the one-call clone_reference_url also create some selection ambiguity, though descriptions provide useful guidance.
Every tool follows a clear verb_noun snake_case convention (e.g., inspect_url, capture_reference_bundle, enqueue_clone_job, detect_runtime_capabilities). The convention is applied uniformly across capture, build, queue, and verification families.
19 tools is a bit heavy but justifiable for a multi-stage cloning pipeline with browser capture, fidelity verification, bundle building, and an async job queue. Each tool maps to a distinct pipeline stage or queue operation, though the surface could be tightened by merging overlapping preflight checks.
The surface covers the full lifecycle from URL inspection and policy auditing through capture, reproduction planning, bundle building, verification, one-call cloning, and job queue management. Minor gaps remain, such as explicit session/auth handling for logged-in captures and job deletion (cancel is the closest analog).