Source: https://www.agentweb.us/comparisons/agentweb-vs-context-dev/

# AgentWeb vs Context.dev: which one lets your agent actually do the thing?

By Andrey Marey, Founder of AgentWeb · As of Oct 20, 2026

> **Short answer.** If your agent has to *do* something on a website, like send a LinkedIn message, place an Uber Eats order or book a table, use AgentWeb. AgentWeb is the execution layer for 1,462 real sites (7,073 actions as of Oct 20, 2026), including the user's own logged-in accounts. Every completed action returns a structured result and a receipt, at $0.05 per completed action, with failed actions refunded. Context.dev is a web scraping API: it is very good at turning public pages into clean Markdown or JSON, but what it hands back is the page, not a finished task.

## At a glance

| | **AgentWeb** | Context.dev |
|---|---|---|
| What it is | Execution layer for AI agents: typed actions on real sites | Web scraping API for AI agents: pages into Markdown, HTML, screenshots or JSON |
| What your agent gets back | A structured result plus a receipt for each action | Page content in the formats you asked for |
| Taking actions (order, book, send, post) | Yes. 2,246 actions tagged write across 1,087 maps | No action endpoints. Browser actions (plain-language "perform" steps and scrolls) run before the page is captured |
| Logged-in accounts | Hosted sign-in. 1,115 maps run on the user's own account. Credentials are vaulted and never shown to the agent | Authenticated browser profiles, on paid plans only |
| LinkedIn | 47 actions: 40 fetch (profiles, people, companies, inbox, posts) and 7 act (message, connection request, InMail, post, invitations) | Read-only: People Enrich accepts a LinkedIn profile URL as an input clue (20 credits per returned match) |
| Coverage model | Per mapped site and action: 1,462 sites, 29 categories, live list at /sites/all | Any URL you can reach, plus crawl (up to 500 pages per call) and batches of up to 25,000 URLs |
| Pricing unit | $0.05 per completed action. Credit blocks of $25, $50 or $100 that never expire. Failed actions refunded | Credits per operation: scrape 1 credit (2 with browser actions), JSON extraction adds 4 |
| Entry price | Free account. read_page free up to 20 reads an hour. Actions run once you add credits | Free plan with 1,000 credits a month. Paid from $19 a month (Developer), then $99, $299 and $499 |
| How agents connect | MCP (https://mcp.agentweb.us/mcp) or REST | REST, SDKs, hosted MCP (mcp.context.dev), ChatGPT plugin |

## What AgentWeb is

AgentWeb is the execution layer for AI agents. Your agent connects once over MCP or REST, then calls typed actions on real sites. On LinkedIn, for example, it can call `get_profile` and `send_connection_request`; on Uber Eats, `add_item_to_cart` and `place_order`; on SevenRooms, `book_reservation`.

When an action needs the user's account, the user connects it through a hosted sign-in. The agent never sees the password. Each action declares its inputs and returns a structured result, so the agent can prove the task finished instead of guessing from a screenshot. Retries are safe with an Idempotency-Key, and you only pay for completed actions.

Coverage is honest and checkable. As of Oct 20, 2026, the live index lists 1,462 sites and 7,073 actions, and any agent can check a domain with `discover` and `list_actions` before it promises anything. If a site is missing, you can request a map.

## What Context.dev is (one fair line)

Context.dev, formerly Brand.dev, is a web scraping API that turns any URL into LLM-ready Markdown, HTML, screenshots or typed JSON. It crawls and batch-scrapes at scale and adds brand, people and news data under one key. If you need lots of public pages read cleanly, it does that job well.

## Where Context.dev falls short for agents that have to act

- **It returns pages, not outcomes.** Every scrape returns the formats you asked for. Even after browser actions run, you get the page back, not a confirmation that the order went through. Your code still has to work out whether it worked.
- **Clicks are prose, not a contract.** Browser actions are plain-language instructions like "Click the product details tab". Nothing declares the inputs or the expected result, so there is nothing to check a finished task against.
- **LinkedIn is read-only.** People Enrich will take a LinkedIn profile URL as a clue and enrich that person (20 credits per match). Its API reference lists no endpoint to message them, connect with them, send an InMail or post. AgentWeb does all four on the user's own account.
- **Logged-in pages cost a plan.** Authenticated browser profiles need a paid plan. On AgentWeb, hosted sign-in comes with every account, and you pay per completed action.
- **You pay for the page, not the outcome.** To be fair, Context.dev doesn't charge for blocked pages or failed monitor runs. But a scrape that loads fine and still doesn't finish your task costs credits. AgentWeb only charges for completed actions and refunds failed ones.

## Which one should you pick?

- **Pick AgentWeb** when the agent has to finish something on a site: order, book, buy, send, post, accept, track. Especially pick it when that happens on the user's own logged-in account and you need proof it happened. That covers most agents people message on iMessage or WhatsApp.
- **Context.dev fits** when the whole job is reading lots of public pages into Markdown or JSON, such as RAG ingestion, monitoring or bulk crawling. AgentWeb has no crawler or bulk batch endpoint.

## Frequently asked questions

### Is AgentWeb a Context.dev alternative?
Yes, when your agent needs to act. AgentWeb is the execution layer on 1,462 real sites (as of Oct 20, 2026), including logged-in accounts, and returns a receipt for each action. Context.dev is the stronger fit if all you need is public pages turned into Markdown at volume.

### Can Context.dev send LinkedIn messages?
Not according to its public API reference as of Oct 20, 2026. Its People Enrich endpoint accepts a LinkedIn profile URL as an input to enrich a person, which is a read. AgentWeb's LinkedIn map includes `send_message`, `start_conversation`, `send_connection_request`, `send_inmail` and `create_post`, run on the user's own account.

### Does AgentWeb read pages too?
Yes. `read_page` returns any accessible public page as structured text and metadata, and it is free up to 20 reads an hour per account. There are also 4,827 typed fetch actions, many of them behind a login, like LinkedIn `get_profile_experience` or Uber Eats `get_active_orders`.

### How do logins work on AgentWeb?
The user connects their account once through a hosted sign-in started by `agentweb_start_connection`. Credentials are vaulted per account and never exposed to the agent. Every action is scoped, permissioned and receipted.

### How is pricing different?
AgentWeb charges $0.05 per completed action, sold in $25, $50 or $100 credit blocks that never expire, and failed actions are refunded. Context.dev charges credits per operation (1 per scrape, 2 with browser actions, plus 4 for JSON extraction), on monthly plans from free (1,000 credits) to $499.

### Which AI tools does AgentWeb work with?
Any MCP client, including Claude, Claude Code, Cursor, Codex and ChatGPT. In Claude Code it is one command: `claude mcp add agentweb --transport http https://mcp.agentweb.us/mcp`.

## Talk it through

Bring me the task your agent keeps failing at and the sites it has to touch. I'll tell you straight whether AgentWeb covers it today. [Book a call](https://cal.com/andreym/agentweb-discovery?utm_source=comparison&utm_medium=organic-content&utm_campaign=agentweb-vs-context-dev)

AgentWeb is not affiliated with Context.dev. Product details change, so test your exact sites and tasks. Corrections: support@agentweb.us

## Sources (accessed Oct 20, 2026)

- AgentWeb coverage (1,462 sites, 7,073 actions, 4,827 fetch and 2,246 act, 1,115 logged-in maps): https://mcp.agentweb.us/api/sites and https://mcp.agentweb.us/maps/library/summary
- AgentWeb pricing, refunds, read_page allowance, MCP and REST: https://www.agentweb.us/pricing , https://www.agentweb.us/llms.txt
- Context.dev definition, credit costs, crawl and batch limits, People Enrich, MCP, plans: https://www.context.dev/llms.txt
- Context.dev scrape API (browser actions, authenticated profiles on paid plans): https://docs.context.dev/api-reference/web-scraping/scrape
- Context.dev People Enrich (LinkedIn URL as input example): https://docs.context.dev/api-reference/people/enrich
- Context.dev pricing: https://www.context.dev/pricing