Your audits, inside your AI assistant.
https://safelaunch.site/mcp
- Transport
- Streamable HTTP (remote)
- Sign-in
- OAuth 2.1, dynamic client registration
- Access
- Read-only · 5 tools
- Price
- Free with every account
Overview
Compliance answers where you already work.
SafeLaunch checks your website, repository or app against 138 privacy, cookie, AI, consumer, accessibility and security laws in 54 markets. The MCP server brings those results to your AI assistant, so it can explain a finding, write the fix and check a sign-off on your behalf.
Connect once
Paste one URL or click Add. Sign in to SafeLaunch and approve the app.
Ask in plain words
“What's critical in my last audit?” Your assistant picks the right tool.
Fix and re-check
Apply the fix prompt, then re-scan in SafeLaunch to update your score.
What you can do
Four things teams use it for.
Triage without leaving your editor
Pull the critical and high findings from your latest audit, with the law behind each one and the file it points to, right next to your code.
“What are the critical findings in my latest SafeLaunch audit?”
Fix with the tool you build in
Fetch a fix prompt written for Lovable, Cursor, Claude Code, Bolt, v0, Replit or Windsurf, ending with checks that prove each fix is done, and let your agent apply it.
“Get the Cursor fix prompt for everything in my last audit, then apply it.”
Track progress across versions
Every audit covers one exact version: a commit, an upload or a dated website capture. Compare scores and see what's still open before you ship.
“List my audits for the checkout project with their scores.”
Check a sign-off before you rely on it
Look up any public verification record: whether it's valid, the exact version it covers, the date it was issued and any accepted exceptions.
“Is sign-off 3f9c… still valid, and which commit does it cover?”
Connect
Set up in under a minute.
Pick your app. Every route signs you in with your SafeLaunch account; nothing to install on your machine.
- 1In Claude (web or desktop), open Settings → Connectors and choose Add custom connector.
- 2Name it SafeLaunch and paste the server URL.
- 3Click Connect, sign in to SafeLaunch and allow access. Then turn the connector on from the tools menu in any chat.
Custom connectors are available on Claude's paid plans; on Team and Enterprise an owner may need to add it first.
https://safelaunch.site/mcp
Check it works
Ask: “Using SafeLaunch, list my audits.” You should see your latest scans with their scores.
Tool reference
Five read-only tools.
Each tool is marked read-only for your client, so most apps run it without extra confirmation. Your assistant chooses them for you; the details are here if you're building an agent.
list_projects
Needs: Any accountYour SafeLaunch projects and the GitHub repository each one is connected to.
Parameters
None.
Returns
id, name, repo, default_branch for each project.
Try asking
“Which SafeLaunch projects do I have?”
Example result
{
"projects": [
{
"id": "8c1e…",
"name": "Checkout app",
"repo": "acme/checkout",
"default_branch": "main"
}
]
}Example workflows
What a session looks like.
Illustrative conversations: your assistant decides which tools to call and in what order.
Fix the serious issues before launch
Cursorlist_audits → get_findings(severity: high)
get_fix_prompt(builder: cursor)
Answer a customer's security questionnaire
Claudelist_audits → get_findings → get_verification
What each plan unlocks
The same access as the website.
The server is free. What a tool returns depends on the report you've unlocked, exactly as on safelaunch.site.
| Tool | Free scan | Launch Report | Launch Fix |
|---|---|---|---|
| list_projects | |||
| list_audits | |||
| get_findings | Findings and actions | Plus before/after code | |
| get_fix_prompt | |||
| get_verification |
See plans and pricing. Upgrading a report costs only the difference.
Security & data
Safe by design.
Permissions
- Signs in with your SafeLaunch account; you approve each app once
- Sees only your own projects, audits and findings (row-level security)
- Reads public sign-off records, which anyone can check
- Can't start scans, buy anything or spend credits
- Can't open pull requests or touch your GitHub repository
- Can't issue, change or revoke sign-offs
Good practice
- Standard OAuth 2.1: SafeLaunch never sees your AI app's password, and your AI app never sees yours. API tokens are stored hashed and shown once.
- Tool results go to the AI app you connected, under its own data terms. Findings quote what we saw on your pages and in your code: treat them as data, not instructions.
- Review the changes your agent makes from a fix prompt, then re-scan. A fix is not a sign-off; only a new audit and a verification record are.
Troubleshooting
If something doesn't work.
“This report hasn't been unlocked yet.”
Findings come from reports you've unlocked. Open the audit in SafeLaunch and choose Launch Report or Launch Fix.
“Fix prompts are included in Launch Fix.”
get_fix_prompt needs a Launch Fix report. Upgrade from the report page; you only pay the difference.
“Audit is running” or “Audit is queued”.
The scan hasn't finished. Most take a minute or two; ask again once the audit shows Complete.
“Audit not found.”
The assistant is signed in as a different SafeLaunch account, or the ID is wrong. Ask it to run list_audits first, or reconnect with the right account.
Unauthorized, 401, or the sign-in loops.
Disconnect SafeLaunch in your AI app and connect again. With an API token, check the header is exactly Authorization: Bearer sl_… and the token hasn't been revoked.
The client can't connect to a remote server.
Older clients only run local servers. Use the mcp-remote bridge shown under Any client, or update the app.
The assistant doesn't use SafeLaunch.
Make sure the connector is turned on for the chat, then mention SafeLaunch by name: “Using SafeLaunch, list my audits.”
Still stuck? Email hello@raghavrhanda.com with the app you're using and the error you see.
Technical details
For agent builders.
- Endpoint (OAuth)
- https://safelaunch.site/mcp
- Endpoint (API token)
- https://safelaunch.site/api/mcp · Authorization: Bearer sl_…
- Transport
- MCP Streamable HTTP
- Auth discovery
- https://safelaunch.site/.well-known/oauth-protected-resource
- Authorization
- OAuth 2.1 with dynamic client registration
- Server
- safelaunch · version 1.0.0
- Tools
- list_projects, list_audits, get_findings, get_fix_prompt, get_verification (all readOnlyHint)
- Results
- JSON text content plus structuredContent (fix prompts return plain text)
- Limits
- list_audits returns your 25 most recent audits; filter by projectId for older projects
Describing SafeLaunch to an assistant rather than connecting one? Use the AI instructions page.
FAQ
MCP, answered.
No audit yet?
Run a free scan first, then ask your assistant what to fix. Connect GitHub to scan private repositories in one click.