Connect your assistant to jwatte.com

The jwatte.com connector helps your assistant find a website tool, make a link to run it, search this site's articles, or look up J.A. Watte's books. It returns information and links. It does not inspect a website, buy a book, send a message, or change anything.

It is listed in the Claude connector directory as a Community connector, which means it passed automated review and has not been verified by Anthropic. You can also add it by its address as a custom connector.

Connection details

Server URL: https://jwatte.com/mcp
Transport: Streamable HTTP. Requests use HTTP POST and receive JSON. Opening the address as an ordinary web page returns a method error; that is expected.
Sign-in: none. No jwatte.com account, API key, OAuth credentials, or payment is needed.
Cost: the connector and website tools are free. The books are sold separately. Your assistant's own plan and usage limits still apply.
Technical details: the server card identifies the endpoint and supported protocol versions.

How to connect

Claude

Open Customize > Connectors.
Select Add custom connector. If a name is requested, enter jwatte.com.
Enter https://jwatte.com/mcp as the server URL. If asked for authentication, choose No sign-in. Leave credentials and request headers empty.
Select Add. Check that it shows Connected and that its Tool permissions list the six tools below.
In a chat, open + > Connectors, turn on this connector, and try one of the requests below.

On Team or Enterprise, an Owner first opens Organization settings > Connectors, selects Add > Custom (then Web if asked), enters the same address with no sign-in, and selects Add. Members then find the Custom connector in Customize > Connectors and select Connect. An Enterprise role with permission to manage organization libraries can also add it.

Claude currently supports custom connectors on its Free and paid plans; Free permits one custom connector. These steps follow Claude's official custom connector guide, checked November 6, 2026.

Claude Code

With Claude Code installed and ready to use, run this in the project where you want the connector:

claude mcp add --transport http jwatte https://jwatte.com/mcp
claude mcp get jwatte

Inside Claude Code, use /mcp to check the server status, then ask for a tool or article. No authorization header is needed. To remove this connection, run claude mcp remove jwatte. See Claude Code's official MCP instructions.

The six tools

Every tool is read-only. Each advertises that it reads data, does not destroy data, and can be called again without changing the site.

find_tool: Give it a need in plain words. It returns matching tools with titles, short descriptions, links, and takes_url, which says whether a run link is available. Optional: category and limit (1 to 10; default 5).
list_audit_tools: Browse the tool catalog. It returns tool slugs, titles, categories, descriptions, links, whether each takes a URL, the total count, and the next page offset. Optional: category (audits, generators, or utilities), keyword search, limit (1 to 20; default 10), and offset. Ask for 3 to 5 results when you only need a shortlist.
audit_url: Give it a tool slug with takes_url: true and a public url. It returns a run_url, the normalized address, tool details, and fetched_by_server: false. Opening the link fills in the address and starts the selected tool. For page audits such as Mega Analyzer, the opened page sends the address to jwatte.com's fetch function, which fetches the target site and may use configured proxies. Results appear on the tool page. This MCP server never fetches or audits the address. You must open the link to get results. Use only a site you own or have permission to check; a link gives no permission for additional checks.
search_blog: Give it a keyword query. It returns matching article titles, publication dates, links, and summaries, not full articles. Optional limit: 1 to 25; default 10.
get_book_info: Give it one book slug: w2trap, 97launch, condotrap, resaletrap, 20agency, or 100network. It returns the title, subtitle, page count, format, listed price, short factual summary, cover, plain Amazon link, and companion website from this site's catalog. It does not check Amazon's current price or make a purchase.
recommend_book: Give it a reading topic. It checks subject phrases and related words for J.A. Watte's six-book catalog and returns one matching title, a factual reason, and links. Generic words such as “launch,” “fire,” or “quote” do not identify a book's subject on their own. These are the publisher's own books; this is not an independent comparison. If no rule matches, it says “No match,” with matched: false, primary: null, and secondary: null. It recommends no book and returns only a catalog link, with no Amazon link.

Try these requests

“Find three jwatte.com tools for checking website accessibility.”
“List five schema tools from jwatte.com's audit catalog.”
“Make a Mega Analyzer run link for https://jwatte.com. Explain what happens when I open it.”
“Find three jwatte.com articles about llms.txt.”
“Show the catalog details for The $97 Launch.”

Limits and errors

The catalog is a published snapshot, not a live scan of websites or shops. Search is keyword matching and can return no matches. Shorter, more specific terms usually help. The server does not read your files, assistant memory, chat history, or other conversations, and cannot submit forms or post reviews.

The host applies a limit of 60 requests per minute per IP address and domain. People behind the same network or assistant service may share that allowance. Requests are limited to 64 KiB and tool execution to five seconds. A rate-limit response means wait about a minute before retrying; do not repeatedly reconnect.

Run links accept public HTTP or HTTPS addresses only. Private or loopback addresses, internal hostnames, embedded credentials, and nonstandard ports are refused. Never include passwords, access tokens, private query strings, or personal health information in tool inputs.

Invalid input: an empty need produces an error naming arguments.need and its minimum length. Supply a short description and try again.
Unknown tool slug: the error says there is no such tool and suggests using find_tool or list_audit_tools.
Refused address: the answer explains why the address is not accepted and says “Nothing was fetched.” Use a public address without credentials.
No search match: an empty result is not an audit failure. Try a different keyword or browse the catalog.
Unexpected server failure: the answer identifies a site-side problem. Retry later or contact us with the tool name and time, without private inputs.

Data handling

The server processes the tool name and arguments your assistant sends, along with the connection request. It uses them to search the site's bundled catalog or form a link. It does not request your full conversation. Text you or your assistant put into an argument is still received, so send only what the tool needs.

The MCP handler has no database or file writes for calls and no persistent conversation or input history. Successful calls do not explicitly write an application log line. Failures can log the tool or method name, configuration or output-validation details, and unexpected error messages or stack traces. We do not promise those diagnostics can never contain an input value.

Netlify hosts the endpoint, processes incoming requests and connection metadata such as IP addresses, applies rate limits, and holds function diagnostics. The handler does not set a deletion schedule for host logs. Netlify's log availability and retention depend on its plan, settings, and policies; we have no confirmed site-specific period to state. See Netlify's privacy policy.

The six tool handlers make no outgoing API requests and do not forward your inputs to Amazon, another website, or an AI provider. The result goes back to your assistant, whose own data policy applies. If you open a returned link, your browser contacts that destination; a run link includes the address you supplied and starts the selected tool. For page audits such as Mega Analyzer, the opened page sends that address to jwatte.com's fetch function, which fetches the target site and may use configured proxies. That separate tool's privacy rules then apply.

The MCP endpoint sets no cookies and runs no browser analytics. This guide and linked website pages use the site's normal cookie and analytics choices. Read the MCP privacy section and cookie policy.

Support and discovery

Use the contact form or email joshua.a.watte@gmail.com, the support address already published by this site. Include the tool name, error text with private details removed, and the UTC time. You review and send any message yourself.

The server is in the official MCP Registry as com.jwatte/site: view the entry. A Registry entry and a Claude directory listing are separate things; neither is needed to connect by URL.

For other agent interfaces, the A2A site-guide card describes a separate service that finds pages on this site. The agent reference explains that interface and browser tools; the setup instructions above are for MCP.

Last updated: November 7, 2026.