MCP Tools Reference
Complete reference for all browser, workflow, integration, AI, and data tools available via MCP
PROMPTERYX_EXECUTION_MODE=local); on the hosted server use browser_ai_action instead.| Tool | Parameters | Description |
|---|---|---|
browser_navigate | url (string) | Navigate to a URL |
browser_click | selector (string) | Click an element |
browser_fill | selector (string), value (string) | Clear and fill an input field |
browser_type | selector (string), text (string) | Type text character by character (for autocomplete, etc.) |
browser_press_key | key (string) | Press a keyboard key (Enter, Tab, Escape, etc.) |
browser_select | selector (string), value (string) | Select a dropdown option by value |
browser_check | selector (string), checked (boolean) | Check or uncheck a checkbox |
browser_hover | selector (string) | Hover over an element |
browser_scroll | direction? (up/down), selector?, pixels? | Scroll the page or to a specific element |
browser_extract_text | selector (string) | Get text content from an element |
browser_extract_attribute | selector (string), attribute (string) | Get an HTML attribute value (href, src, etc.) |
browser_screenshot | fullPage? (boolean) | Take a screenshot (returns image) |
browser_evaluate | script (string) | Execute JavaScript in the page context |
browser_wait | type (selector/timeout/navigation), value | Wait for an element, timeout, or page navigation |
browser_get_page_info | (none) | Get current URL and page title |
browser_drag_drop | sourceSelector, targetSelector | Drag and drop between elements |
browser_set_files | selector (string), files (string[]) | Upload files to a file input |
browser_get_cookies | (none) | Read all cookies for the current page |
browser_set_cookies | cookies (array) | Set cookies |
browser_pdf | path? (string) | Generate a PDF of the current page |
browser_close | (none) | Close the browser session and record usage |
PROMPTERYX_EXECUTION_MODE=local to use your own Chrome with saved logins.| Tool | Parameters | Description |
|---|---|---|
workflow_list | search? (string) | List your saved workflows, optionally filtered by name |
workflow_get | workflowId (string) | Get full details of a specific workflow |
workflow_execute | workflowId (string), variables? (object) | Run a saved workflow with optional input variables |
workflow_create | name (string), description?, nodes? | Create a new workflow from nodes or natural language |
workflow_delete | workflowId (string) | Delete a workflow |
workflow_add_node | workflowId (string), nodeType, params | Add a node to an existing workflow |
workflow_record_actions | actions (array) | Record browser actions and save as a reusable workflow |
Integration tools are dynamically generated based on your connected OAuth services. When you connect a new service on the Connections page, the corresponding tools appear automatically on the next MCP server startup.
More services are added regularly. New integrations appear as MCP tools automatically — no code changes needed.
| Tool | Parameters | Description |
|---|---|---|
browser_ai_action | instruction (string), maxSteps? (default 10), model?, aiVision? (quality preset), cloudProfileId?, persistContext?, viewport?, recordSession? | AI sees the screen and performs clicks, scrolls, typing to complete the instruction |
Defaults to gemini-3.5-flash; pass model for another browser-agent model (for example claude-sonnet-5-5-vision). Spends AI credits and cloud-browser minutes. Access follows your balance on any plan: when credits or minutes run out the tool says which to top up.
| Tool | Parameters | Description |
|---|---|---|
connect_hub_list_apps | search? (string), limit? (1-100, default 25) | Browse or search the catalogue of 3,000 apps; returns app slugs |
connect_hub_list_actions | app (string) | Actions an app offers, each with its parameters |
connect_hub_get_action | app, action (string) | Full parameter schema for one action |
connect_hub_list_accounts | app? (string) | The user's connected accounts (ids for credentialId), plus apps connected through Prompteryx Connections |
connect_hub_run_action | app, action, params (object), credentialId? | Run an action with the user's connected account. Uses one Connect Hub call; asks for confirmation in clients that honour destructive hints |
Typical flow: connect_hub_list_apps → connect_hub_list_actions → connect_hub_get_action → connect_hub_run_action. Runs count against the plan's monthly Connect Hub calls.
| Tool | Parameters | Description |
|---|---|---|
chat_send | message (string), chat_id?, model?, wait_seconds? (max 55) | One turn with the Prompteryx assistant (same tools, models, memory and credits as the web chat). New chats show in your history as "via API". Long work returns status "running" + job_id; changes come back under needs_confirmation |
get_job | job_id (string) | Status and result of a chat turn or video started earlier |
list_models | (none) | Chat models you can pick (id, label, provider) and the default |
list_chats | limit?, before? | Your chats, newest first |
get_chat | chat_id (string), limit? | The visible messages of one chat, with images and files |
list_outputs | limit?, chat_id? | Files, documents, images, video and audio the assistant made, with links |
Chat turns spend your AI credits exactly like the web chat, and follow the same plan and trial rules. Anything that would change something outside the chat (send an email, post, create, delete) is never done by chat_send: it comes back under needs_confirmation with a confirm_token for confirm_action.
| Tool | Parameters | Description |
|---|---|---|
list_connected_apps | (none) | Every app you connected: Prompteryx Connections (Gmail, Calendar, Drive, Slack…), Connect Hub accounts and custom connectors |
app_find_actions | app (string), intent? | What an app can do; each action is marked read (runs now) or write (needs confirmation) |
app_run_action | app, action, props?, account_id? | Reads run now. Writes do NOT run: they return needs_confirmation with a summary and a confirm_token |
confirm_action | confirm_token (string) | Run a change the user approved. Each token works once and expires (30 minutes for app actions) |
Flow: list_connected_apps → app_find_actions → app_run_action. A write returns a summary and a confirm_token; show the user what will happen and call confirm_action only after they agree.
| Tool | Parameters | Description |
|---|---|---|
generate_image | prompt, style?, aspectRatio?, model?, confirm_token? | Image from a description |
generate_video | prompt, seconds?, resolution?, aspect_ratio?, audio?, model?, confirm_token? | Short video clip with sound; poll get_job with video_job_id |
generate_music | prompt, duration_seconds?, instrumental?, lyrics?, model?, confirm_token? | Original song or instrumental |
generate_speech | text, voice?, model?, confirm_token? | Voiceover from a script |
generate_sound_effect | prompt, duration_seconds?, loop?, model?, confirm_token? | Sound effect or ambience (ElevenLabs) |
Call once without confirm_token: you get the price in AI credits and a confirm_token (valid 15 minutes, once). Call again with it to start: the request goes straight to the model you were quoted (no chat model, no extra charge). The finished file comes back with url (view) and download_url (save), both valid for 24 hours; small images also come back inline. Results are saved in a new chat and in list_outputs.
| Tool | Parameters | Description |
|---|---|---|
explore_trending | tab? (trending, new, top-week, top-month, all-time, staff-picks), type?, cursor? | Public Explore posts |
explore_search | q (string), type?, cursor? | Search public Explore posts |
| Tool | Parameters | Description |
|---|---|---|
transform_json | data (object), query (string) | Parse, query, and transform JSON data |
generate_uuid | (none) | Generate a random UUID |
base64_encode | text (string) | Encode text to Base64 |
base64_decode | data (string) | Decode Base64 to text |
format_date | date (string), format (string) | Format dates and timestamps |
math_operation | expression (string) | Evaluate a math expression |