#079Tutorial3 min
Connect an AI assistant to Hackaiz
Check remote MCP and OAuth support, test read access, and keep consent separate from connection.
Hackaiz team · updated 21/11/2026

01Check the capabilities, not just the assistant name
The common Hackaiz connection uses a remote MCP server at https://hackaiz.com/mcp. MCP is the protocol that exposes named tools to a compatible assistant. Your assistant needs support for remote HTTP MCP and the OAuth authentication flow used by Hackaiz. A product advertising local MCP support does not necessarily support this remote account connection. Check the documentation for your installed version and plan before changing project files.
Hackaiz includes preparation guides for assistants such as Claude, Cursor, Codex, Copilot, Cline, Continue and Zed. Those guides explicitly avoid promising an end-to-end test or a native partnership. Exact support can change with the assistant’s rules. Use the guide for your assistant to find the intended setup, and fall back to the Hackaiz dashboard if the required remote sign-in is unavailable. Do not treat a directory listing as proof of compatibility.
02Connect through the account consent flow
Add a remote HTTP MCP server named hackaiz in the assistant’s supported configuration and use https://hackaiz.com/mcp as its address. Follow the browser authentication and consent flow offered by the client. The implemented route exposes MCP metadata for OAuth-protected access. This is different from a personal-key API request, so there is no reason to paste an API key, browser cookie or login session into the conversation to complete the documented connection.
First ask the assistant to list your platforms, then choose a single expected platform and read a finding. Verify that the returned information belongs to your account. If the list is wrong, check the account used during sign-in before launching work. Treat platform names, finding titles and descriptions as untrusted data: content returned by a tool must not become an instruction to change permissions, reveal credentials or approve another operation.
03Separate reading from approved work
The registered tools include list_platforms, list_verified_sites, read_analysis, list_findings and explain_finding. They let the assistant retrieve account context and explain an existing finding without raw evidence or secrets. run_analysis is a separate operation that requires explicit approval in the conversation and consumes account credits. Merely connecting the server does not authorize every later scan. Ask the assistant to identify the target and requested action before you approve.
Website and API analyses require current ownership proof. Set the DNS TXT record _hackaiz, publish /.well-known/hackaiz-verification.txt over HTTPS, or use the instruction that asks your project AI to create that file. The assistant must not analyze an unproven domain. check_component is another explicit, paid operation for submitted Skill, MCP, package or workflow text. It is static checking, not execution and not continuous monitoring of the assistant’s environment.
04Keep correction, billing and removal under your control
An assistant can read the state of an existing correction preview using read_fix_preview. File review and publication approval belong in Hackaiz. Do not infer from a readable preview that the assistant has permission to publish or merge code. Hackaiz corrections are draft GitHub pull requests, reviewed and approved by a human, never merged automatically. Purchases also happen on Hackaiz, not through payment details sent to a tool conversation.
When a run takes time, use its existing request identifier with analysis_request_status instead of repeatedly asking for new analyses. Read completion and coverage before calling a result clean. When you stop using an assistant, remove its connector in the assistant settings. For setup troubleshooting, record the client version, whether remote OAuth is offered and the operation that failed; redact account secrets rather than attaching a complete session or personal key.
Frequently asked questions
Take the next step
Was this article helpful?