The web, for
your agents.

Oassis is web infrastructure for AI agents. Its API can extract page content and documents, discover site URLs, search the web, interact with a browser, run batch extraction and crawling, produce research with sources, and monitor pages on a schedule. The production API base URL is https://api.oassis.dev. Web Tools use /web/v1, Web Tasks use /tasks/v1, and scheduled Agents use /agents/v1.

Use Web Tools for one synchronous operation, such as reading a URL or clicking a button. Use Web Tasks for an asynchronous batch, crawl or research job: start it, keep its job ID, then request status and results. Use Agents for recurring price, page-change or competitor monitoring. A map discovers URLs; a crawl discovers and reads pages; research answers a question using sources.

Call POST /web/v1/scrape or the MCP tool web_scrape with a url and formats, for example {"url":"https://oassis.dev","formats":["markdown"]}. One request returns the requested outputs. You can also supply html instead of a URL. GET /web/v1/scrape describes the operation without executing it or charging for it. See /web-scraping-api for examples.

Web Scrape supports markdown, html, links, elements, controls, accessibility, screenshot, pdf and json. Use selectors with the elements format; use json.prompt or json.schema with the json format. Other options include waiting for a CSS selector, viewport settings, request headers and cookies, screenshot and PDF options, and maxAge for cached content. Use the OpenAPI reference for each format's exact schema. Document conversion and browser-rendered pages have different format constraints.

Call POST /web/v1/map or web_map with a url. Map discovers URLs from sitemaps and can also inspect links on the rendered page. Set includePage to false for sitemap-only discovery. Set limit to cap results, search to keep URLs containing text, includeSubdomains to include subdomains, and includePaths or excludePaths to filter paths. The limit accepts 1 to 5,000 URLs. GET /web/v1/map describes the inputs. Map returns URLs rather than page content.

Call POST /web/v1/search or web_search_exa to search the web through Exa. GET /web/v1/search describes the current search parameters and conditions. Search requires payment through account credits or HTTP x402; it is not included in the keyless Scrape and Map trial. Use Scrape to read a result page after finding its URL.

Call POST /web/v1/session or web_session_open with a url. The response provides a session ID and the requested page outputs; controls are the default format. Keep the session ID for later actions and close the session after finishing. GET /web/v1/session describes the parameters. See /browser-api for the complete session workflow.

Call POST /web/v1/act or web_act with the session ID from Web Session Open and the actions to perform. Actions include navigation, clicking, typing, selecting, pressing keys, scrolling, waiting, going back and going forward. Use the control references returned by the browser to target elements. GET /web/v1/act describes the action schema. Actions operate on an existing browser session.

Call POST /web/v1/session/{id}/close or web_session_close with the session ID. Closing ends the browser session and settles any outstanding browser time. There is no separate closing fee, but pending browser time may still be charged. The API also closes an idle session automatically; close explicitly when your workflow finishes.

Call POST /web/v1/feedback or web_feedback to report whether a result was correct. Feedback is free. GET /web/v1/feedback describes the accepted feedback fields. Include the information requested by that schema so the report can be associated with the relevant result.

Start a batch with POST /tasks/v1/scrape/batch or web_scrape_batch, supplying the URL list and supported output options. Save the returned jobId. Read progress and results with GET /tasks/v1/scrape/batch/{id} or web_batch_status. Stop pending work with POST /tasks/v1/scrape/batch/{id}/cancel; in MCP, call web_batch_status with cancel: true. GET /tasks/v1/scrape/batch describes the start parameters.

Start a crawl with POST /tasks/v1/crawl or web_crawl. Supply the starting URL and set the page limit and path filters to define the scope. Read progress and collected results with GET /tasks/v1/crawl/{id} or web_crawl_status. Cancel pending work with POST /tasks/v1/crawl/{id}/cancel; in MCP, call web_crawl_status with cancel: true. GET /tasks/v1/crawl describes the start parameters. See /web-crawling-api for the difference between Map and Crawl.

Start research with POST /tasks/v1/research or web_task_start. The task runs asynchronously and returns a job ID. Read progress, sources, the result and billing with GET /tasks/v1/research/{id} or web_task_status. Stop pending research with POST /tasks/v1/research/{id}/cancel or web_task_cancel. GET /tasks/v1/research describes the research route and its pricing. The /task-api guide explains research requests, budgets and lifecycle.

No. Status and cancellation endpoints operate on an existing job ID and are free. They require proof that you own the job: use the owning account API key, or the access_token returned for an HTTP x402 task, sent in the Task-Token header for research. Cancelling stops pending work; work already completed or in progress may still incur charges. Keep the original job ID and access credentials.

GET /agents/v1 lists the scheduled agents. Price Watch monitors product prices, Web Watch reports page changes and optional keyword matches, and Competitor Monitor watches a competitor's selected pages. Services run daily or weekly. Scheduled Agents currently require a free account API key and allow up to 10 active services per account. They are HTTP services, not MCP tools.

Read GET /agents/v1/price-watch, GET /agents/v1/web-watch or GET /agents/v1/competitor-monitor for each agent's parameters. Create the service by sending POST to the same path with your account API key. Price Watch reads product URLs; Web Watch watches page content; Competitor Monitor watches selected paths on a domain. The first run is queued for a scheduler tick and is not an immediate synchronous result.

Use GET /agents/v1/services to list your services and GET /agents/v1/services/{id} to read one service, its state and its latest report. Use POST /agents/v1/services/{id}/cancel to stop future scheduled runs. These endpoints require the owning account API key. Cancellation retains the latest report, and a read already in progress may finish.

Yes. The homepage includes separate Web Scrape and Web Map demos with https://oassis.dev prefilled. They call the same MCP tools and share the existing one-time 100-credit keyless trial, identified by the caller's network address. The demo needs no account or API key. Other paid tools and continued usage after the trial need account credits or supported HTTP x402 payment.

Use https://api.oassis.dev/mcp as the MCP server URL. POST /mcp accepts the JSON-RPC initialization handshake, tools/list for tool discovery and tools/call for execution. Keyless access covers the Scrape and Map trial. Configure Authorization: Bearer YOUR_API_KEY to use shared account credits for other tools and continued usage. GET /mcp provides discovery information; GET /.well-known/mcp/server.json provides the server manifest. See /mcp for setup examples.

Send Authorization: Bearer YOUR_API_KEY with an HTTP API request to use account credits. Alternatively, supported paid HTTP operations accept x402: the first unpaid request returns an HTTP 402 payment challenge, then a compatible client pays in USDC on Base and retries. GET describes an operation without running or charging for it; POST executes it. Discovery, task status and cancellation are not purchases of a new execution. See /x402 and /docs for examples.

One credit equals $0.002. A standard page output costs 1 credit; JSON extraction costs 2.5 credits, and a cached output costs 0.1 credit. Sitemap-only Map costs 1 credit, or 1.5 credits when the page is included. Opening a browser session costs 2.5 credits plus requested formats; actions cost 0.25 credit each plus outputs and any outstanding browser time. Batch extraction starts at 1 credit per URL and crawling at 2 credits per page. Research uses a budget. The pricing table on this page lists the current rates and settlement rules.

Call GET /web/v1/account with your API key. It returns available and reserved credits and the recharge URL; checking the account is free. A verified account receives 1,000 welcome credits once. Purchased credits do not expire. Create keys and manage credits in /dashboard/billing. Use the same account key for HTTP and MCP to share the balance.

GET /openapi.json provides the machine-readable HTTP API reference. GET /llms.txt provides a text overview with documentation links. GET /mcp and GET /.well-known/mcp/server.json expose MCP discovery information. Human-readable guides are available at /docs, /web-scraping-api, /web-crawling-api, /browser-api, /task-api, /mcp, /x402 and /agents. GET /robots.txt and GET /sitemap.xml provide crawling instructions and the public site URL list.

No. A page may be unavailable, protected by authentication or anti-bot systems, or changed since an integration was written. Keyless reads respect robots.txt. Inspect the returned errors, metadata and sources before relying on extracted content. A successful HTTP response can still contain tool-level errors. Use browser sessions when your workflow requires permitted interaction before reading a page.