Give it to your Agent
Give Claude, Cursor, Codex, ChatGPT, VS Code, and other AI agents secure access to Context.dev through OAuth.
Connect your AI agent to Context.dev once, then ask it to scrape, search, extract, parse, enrich, or monitor the web directly from chat.
Connect your client
Choose your client and follow the steps below. Use the exact hosted URL, including /mcp.
Claude
Cursor
Codex
ChatGPT desktop
VS Code
Other clients
Claude, Cowork, and Claude Desktop
1
2
3
4
Claude Code
Add the hosted server for your user:
terminal
Start Claude Code, run /mcp, select context, and complete the browser sign-in.
Open your global MCP configuration at ~/.cursor/mcp.json and add:
~/.cursor/mcp.json
Then open Cursor Settings → Tools & MCP:
- Find
context. - Select Authenticate.
- Sign in to Context.dev in the browser window.
- Confirm that Cursor shows the server as connected.
Use .cursor/mcp.json instead if you only want Context.dev available in one project.
Add the hosted server:
terminal
Authenticate:
terminal
Run codex mcp list to verify the connection. In the Codex CLI, use /mcp to inspect the available Context.dev tools.You can also add the server from Settings → MCP servers in the ChatGPT desktop app or Codex IDE extension. Choose Streamable HTTP, enter the hosted URL, save, restart the client, and select Authenticate.
1
2
3
4
Run MCP: Open User Configuration from the Command Palette, or create .vscode/mcp.json for a project-specific connection:
.vscode/mcp.json
Start the server from the inline action above its configuration. VS Code opens a browser for OAuth the first time it connects.
Use a remote Streamable HTTP server with OAuth:
mcp.json
Your client must support remote Streamable HTTP MCP servers and OAuth. When prompted, sign in to Context.dev and approve access.
Verify the connection
Start a new conversation and ask:
Use Context.dev to retrieve the live brand profile for stripe.com. Tell me which Context.dev tool you called.
A working connection calls get-brand or brand-retrieve-unified and returns live data. If the agent answers from memory without making a tool call, confirm that Context.dev is enabled for the conversation.
What your agent can do
The MCP server exposes the callable operations in the public Context.dev API, plus a visual brand card.
| Capability | Tools |
|---|---|
| Parse files | parse-document |
| Scrape pages and sites | web-scrape-html, web-scrape-markdown, web-scrape-images, web-scrape-sitemap, web-crawl, web-screenshot |
| Search and extract | web-search, web-extract |
| Brand intelligence | get-brand, brand-retrieve-unified, web-styleguide, web-fonts |
| Products and classification | brand-ai-product, brand-ai-products, web-naics, web-sic |
| Cache warming | utility-prefetch |
| Monitor management | list-monitors, create-monitor, get-monitor, update-monitor, delete-monitor |
| Monitor runs | run-monitor-now, list-monitor-runs, get-monitor-run, list-account-runs |
| Monitor changes and usage | list-monitor-changes, list-changes, get-change, list-monitor-credit-usage, rotate-monitor-webhook-secret |
Prompts to try
Authentication and permissions
Context.dev uses OAuth 2.0 Authorization Code flow with PKCE. The MCP client receives scoped tokens; it never receives your Context.dev password or API key. The server requests:
api.readfor parsing, scraping, searching, extraction, enrichment, and reading monitor dataapi.writefor creating, updating, deleting, or manually running monitors and rotating webhook secrets
MCP calls use your Context.dev account and consume the same credits as the corresponding API operations. See pricing and monitor credit usage.