Skyvern

Create a browser profile from a persistent browser session or workflow run.

skyvern_browser_profile_create

Create a new browser session to start interacting with websites. Creates a cloud-hosted browser by default with geographic proxy support. This must be called before using any browser tools (navigate, click, extract, etc.). Use local=true for a local Chromium instance. The session persists across tool calls until explicitly closed.

skyvern_browser_session_create

Create a new folder for organizing workflows.

skyvern_folder_create

Create a reusable, versioned workflow from a YAML or JSON definition. For multi-page automations, scheduling, and repeated runs — not one-off trials (use skyvern_run_task for those). The ENTIRE workflow — its title, blocks, and parameters — must be serialized INTO the `definition` string (YAML or JSON). Do NOT pass title, blocks, parameters, or other workflow fields as separate top-level arguments; only definition/format/folder_id/code_only are accepted and any other top-level field is rejected. Minimal JSON definition: {"title": "My workflow", "workflow_definition": {"blocks": [...], "parameters": [...]}}. This creates from a definition — it does not list or search; to browse existing workflows use skyvern_workflow_list. One block per step: "navigation" for actions, "extraction" for data. Do NOT use deprecated "task" type. Omit `code_only` or pass null to use this server's default; organization policy may enforce code-only, making rejection intentional. Call skyvern_block_schema() for block types and schemas. Use {{parameter_key}} for input references. When omitted, run_with defaults to "agent" and code_version defaults to 2 for both JSON and YAML definitions. Pass run_with="code" to opt into cached script execution. Blocks share a browser session automatically. Give every code block a `prompt`: its plain-language goal, shown as the block's Goal in the editor. Code blocks that omit `prompt` are defaulted to prompt="" so they render the current code block editor experience. A code block's `steps` outline is always rebuilt from its code, so do not send one. Leave optional toggles and overrides unset unless the user explicitly asks for them. This applies to workflow-level fields (persist_browser_session, pin_saved_session_ip, extra_http_headers, totp_verification_url, totp_identifier, etc.) AND block-level overrides (max_retries, max_steps_per_run, totp_identifier, complete_criterion, error_code_mapping, continue_on_failure, engine, model, etc.). The schema defaults are intentional; silently flipping them changes behavior the user did not request. Skill instructions can be stored in `workflow_definition.workflow_system_prompt`; see skyvern_workflow_get for the read/preserve flow and format caveats.

skyvern_workflow_create

Create a schedule for an existing workflow: on a cron, on a fixed interval, or once at run_at. Cadence and timezone are validated server-side and errors are surfaced verbatim. Returns 501 if workflow schedules are disabled in this build.

skyvern_schedule_create

Declare one authoritative terminal record; does not interact with the browser. status must be exactly: completed — the stated goal was achieved, including when the goal itself requested a safe stop or termination; terminated — deliberately stopped short of the goal because safety, permission, or impossibility was discovered mid-run; failed — attempted but could not achieve the goal. Optionally include output and reason. If schema is provided, output must validate, including format assertions the runtime recognizes. The response itself is the terminal record. A later call supersedes an earlier one only in the caller's own transcript.

skyvern_finish

Delete a browser profile.

skyvern_browser_profile_delete

Permanently delete a stored credential. This cannot be undone.

skyvern_credential_delete

Delete a folder. By default workflows remain and are removed from the folder.

skyvern_folder_delete

Delete a workflow permanently. Use when you need to remove a workflow that is no longer needed. Requires force=true to prevent accidental deletion.

skyvern_workflow_delete

Delete a workflow schedule. Irreversible — schedule stops firing immediately.

skyvern_schedule_delete

Deploy a new version of a cached script with updated files. Creates a new version that will be used on the next workflow run. File content must be base64-encoded.

skyvern_script_deploy

Disable an active workflow schedule.

skyvern_schedule_disable

Drag an element and drop it onto another. Supports AI intent, CSS/XPath selector, or both for source and target.

skyvern_drag

Enable a paused workflow schedule.

skyvern_schedule_enable

Run JavaScript on the page. Supports await (auto-wrapped in async IIFE). For multi-line await, use an explicit return. Full responses are returned by default; use ``verbosity="summary"`` for an opt-in compact response. The mandatory response-size cap still applies. On the page/CDP route an expression that never settles within the browser action deadline returns a TIMEOUT result. Direct JavaScript evaluation is unavailable in extension mode. The extension returns OP_NOT_ALLOWED, which this tool reports as ACTION_FAILED. Use skyvern_observe, skyvern_get_html, skyvern_find, or skyvern_get_value to inspect the page. Use skyvern_click or skyvern_type to interact. Do not enable User Scripts or reroute the expression through CDP or another injection API. Security: executes in page context — use only with trusted expressions.

skyvern_evaluate

Run JavaScript to read the page AND capture a screenshot in ONE call. Direct JavaScript evaluation is unavailable in extension mode. This tool returns the evaluation failure and can still return screenshot evidence. Use inspection and interaction tools instead. A single "do it and prove it" primitive: your JS returns the scraped values and the tool returns them together with a screenshot of the page as visual proof, so every fact you read is backed by an image without a second tool call. The screenshot is saved to a file path by default (pass inline=true for base64). Supports await (auto-wrapped in async IIFE); for multi-line await use an explicit return. Full responses are returned by default; use ``verbosity="summary"`` for an opt-in compact response. The mandatory response-size cap still applies. Security: JS executes in page context — use only with trusted expressions.

skyvern_evaluate_and_screenshot

Execute browser operations. On stdio, refs persist across calls until the next observe, navigation, or page/document context change. In hosted stateless HTTP, refs from prior requests do not resolve; prefer selector or intent params, using observe+ref steps in one batch only when refs are predictable in advance. Allowed tools: navigate, click, type, press_key, select_option, hover, scroll, wait, observe, screenshot, evaluate.

skyvern_execute

Extract structured data AND capture a screenshot of the page in ONE call. Use this to record a finding together with its visual proof in a single step, instead of a separate skyvern_extract + skyvern_screenshot. The screenshot is saved to a file path by default (pass inline=true for base64) and returned alongside the extracted data, so a reviewer that only credits visible evidence can see it. Full extracted data is returned by default; use ``verbosity="summary"`` for an opt-in compact response. The mandatory response-size cap still applies.

skyvern_extract_and_screenshot

Extract structured data from the current page using AI and a dedicated extraction LLM. Navigate first. Optionally provide a JSON schema to enforce output structure. Full extracted data is returned by default; use ``verbosity="summary"`` for an opt-in compact response. The mandatory response-size cap still applies. For visual-only inspection use ``skyvern_screenshot``.

skyvern_extract

Extract one schema-conformant JSON object from the current page. Uses the same AI extraction path as skyvern_extract, then strictly validates the returned value against schema, including format assertions the runtime recognizes. Returns the validated object only after validation succeeds; schema or output violations return actionable JSON paths. Navigate first. Use session_id or cdp_url to target an existing browser.

skyvern_extract_structured

Find elements using a semantic locator API — by role, text, label, placeholder, alt text, or test ID. Returns match count, text content, and visibility. Use to verify elements exist before interacting.

skyvern_find

Get 1Password configuration metadata for this organization. The service account token is never returned.

skyvern_onepassword_config_get

Get Bitwarden configuration metadata for this organization. The master password is never returned.

skyvern_bitwarden_config_get

Get a browser profile by ID.

skyvern_browser_profile_get

Get details about a specific browser session -- status, timeout, availability. The result includes app_url for the live view and recording_url for the first recording.

skyvern_browser_session_get

Return trajectory_json, entry_count, truncated, and capture_status for a browser session. Pass trajectory_json directly to skyvern_code_block_synthesize. Only deterministic skyvern_click, skyvern_type, skyvern_select_option, and skyvern_press_key actions are captured, so entry_count reflects just those; retrieve before closing the session (close deletes the trajectory). A not_found capture_status means nothing was captured, or the trajectory expired, was evicted, or lives in another server process. Capture is per server process, so hosted multi-replica results may be partial.

skyvern_trajectory_get

Read console log messages from the browser. Filter by level ('error') or text substring.

skyvern_console_messages

Get a stored credential's metadata by ID. Returns name, type, and username — never the password or secret.

skyvern_credential_get

Get HTML content of a DOM element identified by `selector` on the CURRENT page. Returns innerHTML by default; set outer=true for outerHTML. This does NOT accept a url and cannot fetch HTML by URL — to read a page you have not opened yet, navigate there first with skyvern_navigate(url=...), then call this with a selector.

skyvern_get_html

Get computed CSS styles from a DOM element. Specify properties for targeted lookup or omit for all.

skyvern_get_styles

Get the current value of a form input element (<input>, <textarea>, <select>).

skyvern_get_value

Get details for a specific folder.

skyvern_folder_get

Get full details for a network request by request_id: response headers and captured body.

skyvern_network_request_detail

Read captured network requests/responses. Use request_id with skyvern_network_request_detail for headers and body.

skyvern_network_requests

Get the caller's organization settings. Use this to discover valid keys before calling skyvern_org_update.

skyvern_org_get

Read uncaught JavaScript exceptions. Distinct from console.error — use skyvern_console_messages for those.

skyvern_get_errors

Get the generated Python code for a cached script. Returns the main orchestrator script and per-block code. Use skyvern_script_list_for_workflow to find script IDs first.

skyvern_script_get_code

List AI fallback episodes for a workflow's cached scripts. Each episode records when a cached script's selector failed and the AI agent took over. Shows error details, block label, and whether the fallback succeeded. Useful for understanding why a script fell back to AI and how the script evolved.

skyvern_script_fallback_episodes

Read sessionStorage values from the current page. Returns all key-value pairs, or specific keys if provided. Use ``verbosity="full"`` to recover raw values up to the mandatory response-size cap. Useful for reading auth tokens, user preferences, or temporary state stored by web apps.

skyvern_get_session_storage