Spicrawl integrations

Spicrawl runs a hosted MCP server. Any agent that can reach it gets 25 spicrawl_* tools for reading web pages.

How it connects

Every integration uses the same hosted MCP server. Your client needs these settings:

URL:        https://mcp.spicrawl.com/mcp
Transport:  Streamable HTTP
Header:     Authorization: Bearer $SPICRAWL_API_KEY
OAuth:      none (API key only)

Once connected, the client lists 25 spicrawl_* tools. Keep the key in an environment variable, not in a committed file.

Coding agents

Claude Code: One command, a plugin, or a shared .mcp.json.
Cursor: .cursor/mcp.json, plus cloud agents.
Codex: One command, or the spicrawl CLI in a shell.
VS Code (GitHub Copilot): .vscode/mcp.json for agent mode.
Windsurf (Devin Desktop): Devin Local or Cascade.
Gemini CLI: settings.json with httpUrl.
OpenCode: Setup in the docs.
Cline: Setup in the docs.
Kilo Code: Setup in the docs.
Zed: Setup in the docs.
JetBrains: Setup in the docs.
Goose: Setup in the docs.
Kiro: Setup in the docs.
Factory: Setup in the docs.
Devin: Setup in the docs.
OpenHands: Setup in the docs.

Agent frameworks

Give an agent you build the same tools. Each framework connects to the hosted server with a header.

LangChain and LangGraph: langchain-mcp-adapters.
OpenAI Agents SDK: MCPServerStreamableHttp.
Vercel AI SDK: createMCPClient from @ai-sdk/mcp.
CrewAI: MCPServerHTTP on the mcps field.
Google ADK: McpToolset.
Mastra: MCPClient from @mastra/mcp.
Hugging Face smolagents: MCPClient from smolagents[mcp].

Any MCP client

Any client that speaks Streamable HTTP and can send a custom header works. A client that supports only stdio can use the mcp-remote bridge.

Crush, Amp, Kimi Code CLI, Warp, Qwen Code: Setup in the docs.
Continue, Augment, Amazon Q Developer, Google Antigravity: Setup in the docs.
Claude Desktop: Uses the mcp-remote bridge.
App builders: Setup in the docs.

Good to know

No crawling. Spicrawl does not follow links or find pages. Give the agent the URLs.
Credits per request. A plain fetch is 1 credit. JavaScript rendering is 3. Full-browser rendering is 8.
Failures are free. A failed request costs 0 credits.
The cache saves time, not credits. A cache hit costs the same as the fetch that stored it.
Cap each call. Set max_cost, and a request that could cost more is refused before it runs.
Coming soon: managed proxies, stealth mode, a remote browser and AI extraction from a plain-language prompt. Today you can bring your own proxy.

Frequently asked questions

Which tools work with Spicrawl?

Any MCP client that speaks Streamable HTTP and can send a custom header. The docs have setup pages for Claude Code, Cursor, Codex, GitHub Copilot, Windsurf, Gemini CLI and many more, plus agent frameworks such as LangChain and the OpenAI Agents SDK.

Does the MCP server use OAuth?

No. It accepts API keys only. Send Authorization: Bearer with your key. A client that supports only OAuth cannot connect.

My client only supports stdio. Can I still connect?

Yes. Bridge it with npx -y mcp-remote https://mcp.spicrawl.com/mcp and pass the Authorization header through an environment variable. The docs page "More MCP clients" shows the exact command.

Do I need the MCP server, or can I call the API?

Both work. The MCP server is best when a model decides what to fetch. When your own code already knows the URL, a plain call to POST /v1/scrape is simpler and costs no tool-schema tokens.

How much does each page cost?

A plain fetch costs 1 credit, JavaScript rendering 3 and full-browser rendering 8. Failed requests cost 0. A cache hit costs the same as the fetch that stored it. Ask the agent to set max_cost on each request so it never spends more than you expect.