StackShield Now Has an MCP Server: Run Scans and Triage Issues From ChatGPT, Claude and Cursor

StackShield now ships a Model Context Protocol server. Point ChatGPT, Claude, Cursor, VS Code, Codex or Gemini CLI at https://stackshield.io/mcp, sign in, and your assistant can run scans, read findings, resolve issues and write reports against your real account. It is on every plan, including the trial, from today.

Setup for each client is in the MCP documentation. This post is about why we built it and what it is actually useful for.

Why an MCP server

Security scan results have a habit of living in a tab you open on Fridays. The findings are good, the remediation notes are good, and none of it is in the place where you are actually writing code or deciding what to do this week.

MCP fixes the location problem. It is an open standard that lets an AI assistant discover and call a service's tools directly. Instead of copying a scan report into a chat window and asking what it means, the assistant pulls the report itself, reads the failed tests, checks which issues are still open, and answers from the live data. When you tell it a header is fixed, it can trigger a rescan and mark the issue resolved without you leaving the editor.

Every capability is discovered on connection. There is no prompt to paste and no configuration beyond the URL and a sign-in.

What the server exposes

Sixteen tools in five groups, three resources and two prompts. In short:

• Checks. List, inspect, create, update and delete the domains you monitor.
• Scans. List scans, start a new one, and read every individual test result from a finished scan, including the remediation steps.
• Issues. List and filter issues by severity, status, domain or test type; resolve, reopen and bulk resolve them; export the lot as CSV.
• Dashboard. A posture summary: domains, active scans, priority issues, domains at risk.
• Resources. stackshield://checks, stackshield://issues/critical and stackshield://stats, for assistants that prefer to load context up front.
• Prompts. security_review and scan_report, which walk the assistant through a multi-step workflow and produce a structured result.

Each tool is annotated so the assistant knows which are read-only and which change state. The full list with parameters is on the tools section of the docs.

Use cases

These are the workflows we expect to see most, and the ones we designed the tool set around.

Triage a scan before standup

"Run a scan on app.example.com and tell me which findings are unauthenticated and externally reachable."

The assistant calls trigger_scan, polls get_scan until it finishes, then walks through get_scan and get_scan_test for the failures. Because it can read the technical details for each test, it can sort by what an attacker with no account could reach rather than by the severity label. That is the triage order we recommend, and now it is one sentence instead of a spreadsheet.

Fix the finding without leaving the editor

In Cursor, VS Code or Claude Code the assistant already has your codebase open. Ask it to "look at the open issues for staging.example.com and fix the ones that are middleware changes." It pulls the issues, reads the remediation notes for each, edits app/Http/Middleware or config/cors.php, and when you have deployed, triggers a rescan and resolves the issues that now pass. The finding and the fix live in the same conversation.

Close the loop after a deploy

"We just deployed. Rescan every active domain and tell me what changed."

list_checks, one trigger_scan per domain, then a comparison against the previous scan for each. Configuration drift, like a debug flag flipped during an incident and never flipped back, shows up in minutes rather than at the next weekly scan.

Weekly posture report for someone who does not read dashboards

The scan_report prompt takes a domain and produces an executive summary, detailed findings and next steps. The security_review prompt produces a prioritised remediation plan grouped by severity. Either can be pasted into Slack or a client update as is. If you run an agency, "produce a scan report for every client domain" is now a single request.

Bulk housekeeping

"Resolve every low severity header finding on the marketing site with the note 'accepted, static site behind CDN'."

list_issues with a severity and domain filter, then bulk_resolve_issues. The bulk tool refuses to run without an explicit confirmation, so the assistant will show you the list and ask before it changes anything.

Onboard a new domain

"Add api.example.com, scan it daily at 06:00 UTC, and run the first scan now."

create_check with a schedule, then trigger_scan. Plan limits apply inside the tool, so if you are at your domain limit the assistant is told why and can suggest the upgrade rather than failing silently.

Answer "are we exposed to this?"

When a new advisory lands, ask "do any of our domains have an exposed Telescope or Horizon instance?" The assistant reads stackshield://issues/critical and the relevant test results across every check and gives you a yes or no with the domains listed. That used to be a manual click through each domain.

What it will not do

Some boundaries were deliberate.

• No account changes. The server cannot touch billing, team membership or API keys.
• Team scoped. An OAuth connection sees the team you currently have selected in the dashboard; an API key sees the team that owns it. Neither can reach another team's data.
• Confirmation for irreversible actions. delete_check and bulk_resolve_issues require confirm: true, which a well-behaved assistant only sends after checking with you.
• Rate limited. The server has its own rate limiter so a runaway agent loop cannot exhaust your scan allowance in one go.

Authentication is OAuth 2.1 with PKCE for connectors, or a team API key for command-line tools. Access tokens expire after an hour and refresh silently for up to 90 days. You can see and revoke every connected assistant from the API keys page in Team Settings.

Getting connected

Every client is covered in the setup section of the docs, including ChatGPT, Claude, Claude Code, Cursor, VS Code, Codex CLI, Gemini CLI and Windsurf. Cursor and VS Code have one-click install buttons. Everyone else needs the URL and a sign-in.

If you are not a customer yet, the free scan will show you what StackShield finds on your domain, and the 14-day trial includes MCP access from the first minute.