The API, as tools for your agent.
Sparkcrawl speaks the Model Context Protocol. Point an MCP client at the remote server with your API key, or run the stdio package locally — the tools call the same /api/v1 with the same scopes, quotas and balance.
Remote server
One endpoint, Streamable HTTP, authenticated with the API key in the header:
https://sparkcrawl.com/mcp
| property | value |
|---|---|
| Transport | Streamable HTTP, stateless: one JSON-RPC message per POST, answered as application/json. |
| Auth | Authorization: Bearer sc_live_… — without it, 401 with WWW-Authenticate: Bearer. |
| Protocol | 2025-03-26 · 2025-06-18 · 2025-11-25 |
| Limits | Request bodies over 64 KiB answer 413; the key's scopes decide which tools succeed. |
Create the key at /app/api with the scopes your agent needs — scrape, crawl, search, research.
Claude Code
claude mcp add --transport http sparkcrawl https://sparkcrawl.com/mcp \
--header "Authorization: Bearer sc_live_…"Then ask, for example, “scrape https://example.com as markdown”. Run claude mcp list to confirm the server is connected.
Cursor
Add the server to ~/.cursor/mcp.json (or .cursor/mcp.json in a project):
{
"mcpServers": {
"sparkcrawl": {
"url": "https://sparkcrawl.com/mcp",
"headers": { "Authorization": "Bearer sc_live_…" }
}
}
}Claude Desktop
Claude Desktop runs local servers: add the stdio package to claude_desktop_config.json and restart the app.
{
"mcpServers": {
"sparkcrawl": {
"command": "npx",
"args": ["-y", "https://sparkcrawl.com/downloads/sparkcrawl-mcp-0.2.0.tgz"],
"env": { "SPARKCRAWL_API_KEY": "sc_live_…" }
}
}
}Connectors added in claude.ai on the web require OAuth, which Sparkcrawl does not offer yet — use the configuration above.
Stdio package
The same 31 tools as a local process, for any client that launches MCP servers by command. It needs Node.js and reads the key from SPARKCRAWL_API_KEY; requests go to https://sparkcrawl.com.
SPARKCRAWL_API_KEY="sc_live_…" npx -y https://sparkcrawl.com/downloads/sparkcrawl-mcp-0.2.0.tgzTools
31 tools, one per API capability: scrape, batch, crawl, map, search, research and research_status, plus job tools (get_job, wait_job, list_jobs, export_job, cancel_job), sessions, schedules, artifacts, usage, billing and pricing. Each tool is billed exactly like its API call — see Billing & errors.
research reserves about US$2.50 at its default budget (≈US$25 at maxCostUsd 5), more than the US$1 welcome balance: top up first, or pass a smaller maxCostUsd (from 0.05) for a shorter report.
curl -X POST "https://sparkcrawl.com/mcp" \
-H "Authorization: Bearer $SPARKCRAWL_API_KEY" \
-H "Content-Type: application/json" \
-H "Accept: application/json, text/event-stream" \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'Billing & errors
Every organization has a single balance in US dollars. The Playground, the API and the MCP server all draw from it at the same tariff, and you only pay for pages that succeed.
API reference
Every operation under https://sparkcrawl.com/api/v1, generated from the OpenAPI 3.1 document the server publishes.