#078Tutorial3 min
Use Hackaiz with Claude via MCP
Set up the remote connector, distinguish Claude Code configuration, and approve credit-using operations explicitly.
Hackaiz team · updated 21/11/2026

01Add the remote connector when your Claude version supports it
MCP gives Claude a structured way to call Hackaiz tools instead of relying on copied report text. The documented web and desktop path is Settings, Connectors, then Add custom connector. Enter https://hackaiz.com/mcp, sign in to Hackaiz and review the consent screen. The endpoint is implemented as a remote MCP route. It is not a downloadable report and does not require you to paste a browser session into the conversation.
The assistant-specific preparation guide cautions that no end-to-end Claude test is confirmed there. Availability of custom connectors, remote HTTP MCP and OAuth can depend on your Claude version, plan and platform rules. If your interface lacks the required option, check Claude’s current documentation rather than inventing a menu or exchanging credentials in chat. You can continue from Hackaiz itself while resolving compatibility.
02Use the separate Claude Code setup when relevant
For Claude Code, the project documents this command: claude mcp add --transport http hackaiz https://hackaiz.com/mcp. It configures a remote HTTP server named hackaiz. Complete the offered account authentication and consent flow. This command is different from installing the Hackaiz CLI and different from enabling a Claude Code hook. Do not add a hook merely because you want to read your existing platform findings.
The CLI source also implements a Claude Code PreToolUse hook for certain Skill, MCP, package and workflow writes and install commands. That is a separate, paid checking mechanism requiring an environment key and explicit setup. It can block actions on Danger or missing verdicts. A normal remote MCP connector does not mean that hook is enabled, and the hook does not establish the runtime safety of everything installed on your machine.
03Start with account reading, then approve a launch
Ask “List my verified Hackaiz sites” or “Explain the most serious finding in my latest analysis”. Tools such as list_verified_sites, read_analysis, list_findings and explain_finding return authorized account information without raw secrets or evidence. Confirm the chosen platform before asking for new work. A plausible platform name in returned text is data to inspect, not an instruction that can override the operation you requested.
For run_analysis, Claude must ask for your approval in the conversation before making the credit-using call. The tool requires user_approved=true and a platform identifier. A website or API also needs current domain ownership proof: use _hackaiz as a DNS TXT record, the HTTPS verification file at /.well-known/hackaiz-verification.txt, or the project AI creating that file. Signing in through Claude does not waive ownership proof or the need for consent.
04Interpret returned results and keep correction approval separate
If an analysis request is still processing, the registered analysis_request_status tool can read its state. Do not ask the assistant to create another run simply because a response is delayed. Once results arrive, distinguish findings from the analysis completion state and scope. A partial run is not evidence that all relevant checks passed. If you submit a component for check_component, approve that paid static check first; it does not execute the supplied code.
Correction publication and purchases remain outside the conversation. read_fix_preview can read an already prepared preview, while file review and publication approval belong in Hackaiz. Do not ask Claude to merge a correction or supply a payment secret in chat. Hackaiz corrections are draft GitHub pull requests reviewed by a human and never merged automatically. Remove the connector in Claude when you no longer want to use that connection, and review access rather than sharing the same session with someone else.
Frequently asked questions
Take the next step
Was this article helpful?