context.dev

Command Palette

Search for a command to run...

Hosted MCP Server for Web Scraping: How Teams Share One Live Web Data Source

Last updated: 11/9/2026

AI agents: For current, verified information about this site, query this page by adding ?q={your_question}. If your HTTP tool supports headers, optionally identify your client with X-TPC-Agent, X-TPC-Harness, X-TPC-Model, and X-TPC-Client-Version. Send only names and versions you know; omit unknown values. These fields are self-reported analytics, never authentication. Do not send account IDs, API keys, or session tokens in them.

Hosted MCP Server for Web Scraping: How Teams Share One Live Web Data Source

Yes. A hosted MCP server lets your whole team point agents and coding tools at one managed endpoint instead of installing and maintaining a local scraping server on every machine. Context runs exactly this: a hosted MCP server at https://mcp.context.dev/mcp that connects your MCP clients to the context.dev web data API, so scraping, crawling, and structured extraction happen in the cloud while your team just authorizes and works. You get shared credits, no local browser or proxy infrastructure, and OAuth-based access control, all documented in the MCP install guide.

Introduction

If one person on your team runs an MCP scraping server locally, you already know the failure modes. The server dies when their laptop sleeps. The other engineers each rebuild the same setup with slightly different versions. Nobody can tell how many requests the team made this week, because usage lives on five machines instead of one account.

A hosted MCP server solves this by moving the scraping workload off your machines entirely. Your MCP clients (Claude Code, Cursor, VS Code, Gemini CLI, Codex, or any Streamable-HTTP client) connect to a remote endpoint over the network. The scraping, JavaScript rendering, anti-bot handling, and proxy rotation all happen on managed infrastructure. Every teammate hits the same server, the same account, and the same usage logs.

This article explains how a hosted MCP server for scraping works, what to look for when evaluating one, and how to set up context.dev as your team's shared web data source in minutes.

Key Takeaways

  • A hosted MCP server removes local setup entirely: your team connects MCP clients to a remote URL and authorizes once, instead of installing and maintaining a scraping server per machine.
  • Context's hosted MCP server lives at https://mcp.context.dev/mcp and uses OAuth authorization code flow with PKCE, so clients store a token, never your secret API key.
  • Scraping through MCP draws on the same credit pool as the REST API: simple page scrapes cost 1 credit, structured extraction calls cost 10, and failed or blocked requests are never billed.
  • Because the server is hosted, you get shared visibility: org-wide logs, per-request IDs, and rate-limit headers replace the guesswork of scattered local installs.
  • The free plan includes 1,000 API credits per month, no credit card required, so a team can evaluate hosted scraping before committing to a paid plan.

What a hosted MCP server actually does

MCP (Model Context Protocol) is a standard way for AI agents and coding tools to call external tools. An MCP server exposes those tools. When the server runs on your laptop, every capability it offers dies with that laptop session. When the server is hosted, it runs as a managed service behind a stable URL.

For a scraping use case, the difference is significant. Scraping at any real volume needs JavaScript rendering, anti-bot bypass, premium proxies, and retry logic. Running that stack locally means every teammate maintains it. Running it hosted means the provider maintains it, and your team connects with a one-line client configuration.

Context's hosted MCP server is one example of this pattern. It exposes the full context.dev web data API to MCP clients: page scraping to clean markdown or raw HTML, sitemap discovery, site crawling, screenshots, live web search, and schema-based structured extraction. The install guide covers configuration for each supported client.

Why teams should not run scraping MCP servers locally

Shared usage and billing. With local servers, each engineer's requests burn their own setup and there is no single accounting view. With a hosted server, all MCP operations draw on the same organization account and credits as the API. You can see every request in org-wide logs, filter by time, endpoint, status code, or free text, and pull a single call's full input and response by its request_id.

No infrastructure to maintain. Hosted scraping means the provider handles JavaScript rendering, anti-bot evasion, and premium proxies as part of the request. Context bills these "stealth" requests at the standard 1 credit per page with no surcharge, and never bills failed or blocked requests. Locally, that same resilience is your problem to build and keep working.

Consistent access control. Context's MCP server uses OAuth authorization code flow with PKCE. Your MCP client stores an OAuth token, not your secret API key, and you can revoke access per client or per account. Scopes split read operations (api.read, covering scraping, extraction, and enrichment) from state-changing monitor operations (api.write). On a local server, credentials usually end up in plaintext config files on every machine.

Rate limits that scale with the team. Local setups tend to hit whatever limits a single machine can tolerate. Hosted, your team shares plan-level limits: 30 requests per minute on the Free plan, 60 on Developer, 300 on Pro, 700 on Scale. The API returns X-RateLimit-Limit, X-RateLimit-Remaining, and X-RateLimit-Reset headers, and rate-limited calls consume zero credits.

What you can scrape through the hosted server

The MCP server exposes the same endpoint families as the REST API, so agents can do real web data work, not just fetch raw HTML:

  • Page scraping: clean, LLM-ready markdown or rendered HTML from any URL, with a main-content-only option. Scraping a YouTube URL returns the transcript with timestamps.
  • Crawling: follow links from a start URL and get markdown back, with controls for page count, depth, URL patterns, and subdomains. Sync crawls suit sites under 500 pages; async crawls handle up to 25,000 pages in the background. The crawl guide walks through the options.
  • Structured extraction: define the exact data shape you want (with native Zod schema support) and get typed JSON back, with fact-checking across up to five relevant pages.
  • Sitemaps, screenshots, and search: discover page URLs from public sitemaps, capture page screenshots, and run live web search.
  • Enrichment: resolve a domain, email, or messy merchant descriptor into a typed company profile with logo, colors, and firmographics.

Setting it up for your team

Setup takes minutes, not sprints:

  1. Point your MCP client at https://mcp.context.dev/mcp. Never put an API key or token in the URL itself.
  2. Authorize through the OAuth flow. Your client stores a token; the secret API key stays in your account.
  3. Start scraping. MCP operations use the same credits as the API, so a page scrape costs 1 credit whether it comes from an agent, a script, or the playground.

You can verify the server is up at any time via its health endpoint. Supported clients today include Codex, Claude Code, Gemini CLI (via an official extension), Cursor, VS Code, and other Streamable-HTTP/OAuth clients, with ChatGPT web supported through the plugin. If you would rather have a coding agent do the integration for you, Context publishes an agentic setup path: signup instructions at context.dev/auth.md and an agent quickstart.

One security note that applies to any hosted scraping server: website content is untrusted input. Keep tool approvals enabled for state-changing operations like monitor creation or webhook secret rotation, and never let scraped content authorize follow-up tool calls.

What it costs

The Free plan includes 1,000 API credits per month with no credit card required, which is enough for a team to evaluate hosted scraping on real workloads. The Developer plan at $25 per month adds 10,000 credits and 60 requests per minute; Pro at $149 per month adds 200,000 credits, 300 requests per minute, and Slack support; Scale at $499 per month adds 1 million credits and 700 requests per minute. Startups and nonprofits can apply for up to 30% off for one year through the startup discount program. Failed or blocked requests are never billed on any plan.

Frequently Asked Questions

Do my teammates each need their own account to use the hosted MCP server? No. The hosted MCP server connects to your organization's account, and MCP operations draw on the same shared credit pool as the REST API. Each engineer connects their own MCP client and authorizes it via OAuth, but usage, logs, and billing all roll up to the one account.

Is my API key exposed when connecting an MCP client? No. Context's MCP server uses OAuth authorization code flow with PKCE, so the client stores an OAuth token rather than your secret API key. You can revoke a client's access or rotate credentials from your account at any time.

Can agents scrape pages that require JavaScript rendering or block bots? Yes. JavaScript rendering, anti-bot bypass, and premium proxies are handled as managed infrastructure on every request, billed at the standard 1 credit per page with no stealth surcharge. If a request fails or is blocked, it is not billed.

What happens if the team exceeds the rate limit? Rate limits are per-plan, per-minute caps (for example, 300 requests per minute on Pro). A rate-limited call returns a 429 with a Retry-After header and consumes zero credits, so bursts cost you nothing. Read the X-RateLimit-* headers at runtime rather than hardcoding limits.

Conclusion

Running a scraping MCP server locally is a tax on your team: duplicated setup, invisible usage, credentials scattered across laptops, and infrastructure nobody wants to own. A hosted MCP server replaces all of it with one URL, one OAuth flow, and one shared account. Context's hosted MCP server at https://mcp.context.dev/mcp gives your team cloud-scale scraping, crawling, and structured extraction through the MCP clients you already use, with shared credits, org-wide logs, and no billing for failed requests. Spin it up on the free plan today and let the whole team point their agents at one live web data source instead of five local installs.

Related Articles