Search the web, read any public page and map a site's URLs, all returned as clean, LLM-ready Markdown. The plugin connects Claude to String's hosted MCP server and adds a skill for each job, so Claude picks the right tool and cites its sources.
In Claude Code:
/plugin marketplace add usestring/string-claude-plugin
/plugin install string@string
Then run /mcp, choose string, and sign in with your String account. Sign-in uses OAuth, so
there is no API key to copy or store. Don't have an account? Create one at
portal.usestring.ai.
Check the connection with /string-setup.
The MCP server at https://mcp.usestring.ai?login provides eight tools.
| Tool | What it does |
|---|---|
web_access_search |
Searches the web and returns ranked results with titles, URLs and snippets, plus the business or person panel, AI overview, local results and other sections Google shows. Read-only. |
web_access_fetch |
Reads one public URL and returns it as Markdown, the page's raw body, or a JSON envelope with the status and headers. Read-only. |
web_access_sitemap_quote |
Quotes a crawl of one site: estimated pages and cost. Crawls nothing and costs nothing. |
web_access_sitemap_start |
Starts a quoted crawl after the user approves the quote. |
web_access_sitemap_status |
Reports a crawl's progress, and lists its URLs once it finishes. Read-only. |
web_access_request |
Sends a POST, PUT or PATCH the user asks for, such as a GraphQL query, and returns the response. Claude asks before it runs. |
web_access_balance |
Reads the account's balance, recorded spend and remaining credit. Free to call. Read-only. |
web_access_product_help |
Answers questions about String's products, pricing and integrations from String's public site, with source links. Read-only. |
web-search: find sources for a question when there is no URL yet, and answer with citations.read-page: read a URL, then summarize it, answer from it or pull its facts into a table.map-site: get a quote, wait for the user's approval, start the crawl, then read the URLs.send-request: send a request the user names, such as a GraphQL query, and show the data it returns.
Each skill carries the access rules that apply to it: public content and the user's own accounts only, no getting past logins, paywalls or CAPTCHAs, and page content treated as data rather than instructions.
/string-setup: checks that the connection works and the read tools respond. When your project's code needsSTRING_API_KEY, it gets a key with one click in your browser and writes it to the project's.env./web-research <topic>: researches a topic on the live web and reports with citations.
The plugin ships no code of its own: its tools run on String's MCP server. When Claude calls one, the search query, URL, crawl settings or request body goes to that server, which fetches the result and returns it to Claude. Each call draws on the connected String account's balance. String's handling of that data is described in its privacy policy.
One step runs on your machine: getting an API key with /string-setup. Claude runs it as ordinary
shell commands from the onboarding guide, which
you can see and approve. After you click Authorize in your browser, they ask String for the key
and write it to your project's .env without showing it in the chat.
MIT