> ## Documentation Index > Fetch the complete documentation index at: https://docs.anchorbrowser.io/llms.txt > Use this file to discover all available pages before exploring further. > ## Agent Instructions > Anchor Browser provides cloud browsers for AI agents and automation: stealth browsing with residential proxies, managed authentication into third-party web apps, and browser sessions that run reliably at scale. AI agents that need Anchor credentials should start with https://docs.anchorbrowser.io/quickstart/agent-access. This flow lets agents obtain an API key programmatically without creating a dashboard account. Read that page before attempting authentication or API access. # Perform Web Task > Start from a URL and perform the given task. ## OpenAPI ````yaml /openapi.yaml post /v1/tools/perform-web-task openapi: 3.1.0 info: title: AnchorBrowser API version: 1.0.0 description: > Anchor Browser ([anchorbrowser.io](https://anchorbrowser.io)) provides cloud browsers built for AI agents and automation. Use this API to launch and control browser sessions, run natural-language web tasks, authenticate into third-party web apps with managed identities, and capture recordings — reliably and at scale, without operating browser infrastructure yourself. ## Core resources - **Sessions** (`/v1/sessions`): launch configurable cloud browsers — residential proxies, stealth browsing, captcha solving, profiles, extensions, recording — and connect over CDP with Playwright, Puppeteer, or any CDP client. - **AI tools** (`/v1/tools/perform-web-task`): state a goal in natural language and the built-in agent completes it inside the browser. Related tools fetch webpages, take screenshots, render PDFs, and execute code in a session. - **Tasks** (`/v1/task`): create, version, deploy, and monitor repeatable browser automations built to run reliably in production. - **Applications and identities** (`/v1/applications`, `/v1/identities`): managed authentication into third-party web apps — including multi-step logins and TOTP-based 2FA — so agents can act on behalf of users. - **Profiles** (`/v1/profiles`): persist cookies and login state across sessions. - **OS-level control** (`/v1/sessions/{sessionId}/mouse`, `/keyboard`, `/clipboard`): raw input for flows that resist DOM automation. - Plus batch sessions for running browser fleets in parallel, session recordings, webhooks, events, extensions, client certificates, and billing usage. ## Authentication Pass an API key in the `anchor-api-key` header on every request. Create keys at [app.anchorbrowser.io](https://app.anchorbrowser.io) under API Access. AI agents that have no key yet can obtain one autonomously through the unauthenticated [agent-access flow](https://docs.anchorbrowser.io/quickstart/agent-access) (`POST /v1/agent-access`). ## For AI agents and LLMs - Docs: [docs.anchorbrowser.io](https://docs.anchorbrowser.io), with a machine-readable index at [/llms.txt](https://docs.anchorbrowser.io/llms.txt) and the full corpus at [/llms-full.txt](https://docs.anchorbrowser.io/llms-full.txt). - MCP server for hosted browser control: `https://api.anchorbrowser.io/mcp`. - Official SDKs: `anchorbrowser` on [npm](https://www.npmjs.com/package/anchorbrowser) and [PyPI](https://pypi.org/project/anchorbrowser/). `DELETE` operations terminate sessions or permanently remove resources (profiles, identities, extensions); `GET` operations are safe and read-only. servers: - url: https://api.anchorbrowser.io description: API server security: [] paths: /v1/tools/perform-web-task: post: tags: - AI Tools summary: Perform Web Task description: Start from a URL and perform the given task. parameters: - in: query name: sessionId schema: type: string title: Browser Session description: >- An optional browser session identifier to reference an existing running browser sessions. When passed, the tool will be executed on the provided browser session. requestBody: required: true content: application/json: schema: $ref: '#/components/schemas/PerformWebTaskRequestSchema' responses: '200': description: The result of the autonomous task. content: application/json: schema: $ref: '#/components/schemas/PerformWebTaskResponseSchema' examples: sync: summary: Synchronous response value: data: result: result: nodes_cpu_usage: - node: pool-e1ro5g0nq-559g5 cluster: do-nyc1-demo-infra cpu_avg_percentage: 8.29 - node: pool-e1ro5g0nq-559gk cluster: do-nyc1-demo-infra cpu_avg_percentage: 24.8 async: summary: Asynchronous response value: data: status: running workflow_id: >- perform-web-task-execution-123e4567-e89b-12d3-a456-426614174000-550e8400-e29b-41d4-a716-446655440000 '400': description: Invalid request. content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' '500': description: Internal server error. content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' security: - api_key_header: [] components: schemas: PerformWebTaskRequestSchema: type: object required: - prompt properties: url: type: string description: >- The URL of the webpage. If not provided, the tool will use the current page in the session. prompt: type: string description: The task to be autonomously completed. agent: type: string description: The AI agent to use for task completion. Defaults to browser-use. enum: - browser-use - openai-cua - gemini-computer-use - anthropic-cua - yutori provider: type: string description: The AI provider to use for task completion. enum: - openai - gemini - groq - azure - xai model: type: string description: >- The specific model to use for task completion. see our [models](/agentic-browser-control/ai-task-completion#available-models) page for more information. detect_elements: type: boolean description: >- Enable element detection for better interaction accuracy. Improves the agent's ability to identify and interact with UI elements. human_intervention: type: boolean description: >- Allow human intervention during task execution. When enabled, the agent can request human input for ambiguous situations. max_steps: type: integer description: >- Maximum number of steps the agent can take to complete the task. Defaults to 200. llm_timeout: type: integer minimum: 10 maximum: 600 description: >- Timeout in seconds for each LLM call made by the agent (browser-use agent only). Raise it when tasks with deep context (long histories, large pages) hit LLM call timeouts. When omitted, the agent uses its model-based defaults. secret_values: type: object additionalProperties: type: string description: >- Secret values to pass to the agent for secure credential handling. Keys and values are passed as environment variables to the agent. highlight_elements: type: boolean description: >- Whether to highlight elements during task execution for better visibility. output_schema: type: object description: JSON Schema defining the expected structure of the output data. async: type: boolean description: >- Whether to run the task asynchronously. If true, the task will be run asynchronously and the response will include a workflow ID. default: false PerformWebTaskResponseSchema: type: object properties: data: oneOf: - $ref: '#/components/schemas/PerformWebTaskSyncResponseData' - $ref: '#/components/schemas/PerformWebTaskAsyncResponseData' x-oneOf-labels: - sync - async ErrorResponse: type: object properties: error: type: object properties: code: type: integer message: type: string PerformWebTaskSyncResponseData: type: object title: sync properties: result: type: object description: The outcome or answer produced by the autonomous task. required: - result PerformWebTaskAsyncResponseData: type: object title: async properties: status: type: string enum: - running description: The status of the asynchronous task execution. workflow_id: type: string description: The workflow ID for tracking the asynchronous task execution. required: - status - workflow_id securitySchemes: api_key_header: type: apiKey in: header name: anchor-api-key description: > API key for your Anchor Browser account, sent as the `anchor-api-key` header on every request. Create one at [app.anchorbrowser.io](https://app.anchorbrowser.io) (API Access), or obtain one programmatically via the unauthenticated agent-access flow (`POST /v1/agent-access`). ````