Skip to Content
Getting StartedMCP quickstart (5 min)

MCP quickstart

Connect any MCP-compatible AI client to Scrapewise. After this, Claude can list your scrapers, run them, and read scraped data — by asking in natural language.

Step 1 — Mint an API key with an MCP-capable scope

In the portal: Settings → API Keys → Create new key.

For the Scope field, pick:

  • LLM_READ — read-only access. The agent can list scrapers, view schemas, and read scraped data, but cannot create/run/delete anything. Recommended for the first time you connect Claude.
  • LLM_FULL — read + write. The agent can create scrapers, run them, delete data. Use when you want Claude to take actions on your behalf.

Avoid USER for MCP — USER-scope keys are explicitly rejected by the MCP gateway (returns 403 scope_rejected).

Copy the key. It looks like sw_live_abc1234.<long-secret>.

Step 2 — Connect your MCP client

The MCP gateway is at https://mcp.scrapewise.ai. Three clients walked through below.

Claude Desktop

Open Claude Desktop → Settings → Developer → Edit Config. Add this to the mcpServers block (creating the block if it’s not there):

{ "mcpServers": { "scrapewise": { "url": "https://mcp.scrapewise.ai", "headers": { "Authorization": "Bearer sw_live_abc1234.your-secret-here" } } } }

Replace sw_live_abc1234.your-secret-here with the key you minted in Step 1.

Quit Claude Desktop (Cmd-Q / fully close — not just the window) and reopen. In the conversation input, type / and you should see Scrapewise tools available.

Claude Code

One command:

claude mcp add scrapewise --transport http https://mcp.scrapewise.ai \ --header 'Authorization: Bearer sw_live_abc1234.your-secret-here'

Verify with claude mcp listscrapewise should appear with status .

claude.ai (web)

Sign in to claude.ai  → click your initials (top-right) → Settings → Connectors → Add Custom Connector.

Fill in:

FieldValue
NameScrapewise
URLhttps://mcp.scrapewise.ai
AuthenticationOAuth — NO — use header
Custom headerName: Authorization, Value: Bearer sw_live_abc1234.your-secret-here

Save. In a new conversation, the Scrapewise tools appear in the tool list.

Step 3 — Try it

In any of the three clients, ask:

“List my scrapewise scrapers”

or

“Scrape https://example.com  and tell me what fields are visible”

Claude picks the right tool, calls it, and shows you the result.

What tools are exposed?

Every REST endpoint under the MCP-tag in the OpenAPI spec  is surfaced as an MCP tool. The current set covers:

Tool familyTools
API keyswhoami, generate_api_key, list_api_keys, delete_api_key
Scraperslist_scrapers, get_scraper, create_scraper, update_scraper, delete_scraper, run_scraper, cancel_scraper_job, stream_job_status
Groupslist_scraper_groups, create_scraper_group, run_group, etc.
Datalist_scraper_data, delete_scraper_data, export_scraper_data
Previewpreview_scraper_from_url, preview_scraper_from_curl
Schemaslist_schemas, create_schema, update_schema, delete_schema

For the full tool catalog with input/output schemas, see MCP tools.

Permissions

The agent inherits your key’s scope. With an LLM_READ key, it can only call read-only tools; with LLM_FULL it can call destructive tools too. There’s no separate consent prompt for each tool call — if you don’t want Claude to delete scrapers, give it an LLM_READ key.

For more on scope semantics, see Scopes.

Troubleshooting

  • 403 scope_rejected from any tool call — your key’s scope can’t call this tool. Mint a higher-scope key.
  • 401 Unauthorized — bad Authorization header. Check the Bearer prefix.
  • Tools don’t show up in Claude Desktop — fully quit and reopen (Cmd-Q, not just close window). Verify the config JSON is valid.
  • 429 Too Many Requests — you’re rate-limited. Wait the Retry-After seconds.
  • Tool result contains prompt-injection content — Scrapewise marks scraped content with type: "scraped" in the response envelope. See Prompt injection for the agent-side defense.

What’s next