Use Slydog with Claude
Works with Claude and any other MCP client
Ask Claude about your flags and change them in plain language:
- "Is the nap reminder on in production?"
- "What would home h-42 get for nap-reminder in production?"
- "Roll new-onboarding out to 25% of homes in development."
- "Turn off dark-mode in production."
- "Who turned off checkout-v2, and when?"
Claude acts as you, with your role in each organization. A viewer can look but not change anything, and an editor can't do what needs an admin. It never sees SDK keys or API tokens.
Claude on the web, desktop and mobile
- In Claude, open Settings → Connectors and add a custom connector.
- For its URL, use
https://api.slydog.dev/mcp. - A Slydog window opens. Sign in with GitHub or Google, the same account you use for Slydog.
- Check where allowing sends you. For Claude it's claude.ai. Any app can call itself "Claude", so only allow it if that's where you started connecting. The page lists the organizations Claude will act in, with your role in each.
- Choose Allow, then ask Claude about your flags.
The connector URL is also on the Your account page of the Slydog admin, with a copy button. Disconnect at any time from Claude's settings.
Claude Code
Add Slydog by its URL:
claude mcp add --transport http slydog https://api.slydog.dev/mcp
Then run /mcp in Claude Code, pick slydog and sign in, the
same way as above. The command is also on the Your account page of the
Slydog admin.
The connector doesn't take API tokens: they last a long time and would sit in Claude Code's settings. Scripts and CI use an API token with the management API instead.
Self-hosted
Your server serves the same endpoint at https://<your server>/mcp.
Signing in from Claude needs sign-in turned on and the server on https. Then the endpoint
takes sign-ins only, like the hosted one. Without sign-in, or on plain http except on
localhost, it takes API tokens instead.
How changes work
- You see every change first. Claude shows what the flag does now and what it would do after, and saves only once you agree. If someone else changed the flag in between, nothing is saved and Claude reads it again.
- Production says so. Changes to a protected environment come with a warning that they reach live users as soon as they're saved.
- Approvals still apply. When your organization requires approval for production, Claude's change becomes a request that waits for a second person. Claude tells you so and links to it in the admin. Switching a flag off skips approval when your organization allows it, as it does by default.
- The audit log shows it. Each change is recorded as you, marked "via MCP".
- Apps get changes within seconds, like changes made in the admin.
What Claude can do
| Look things up | Change things (editor or higher) |
|---|---|
| List your organizations, apps and environments | Turn a flag on or off in one environment |
| List flags, with whether each is on in every environment | Serve everyone one variation, or a percentage split |
| Read a flag's rules, in words | Add or remove specific targets, such as home h-42 |
| Check what a flag gives one user, home or device | Replace a flag's rules |
| List segments and who's in them | Create or edit a segment |
| Read the audit log | Create a flag, switched off everywhere |
Keys, tokens, members, environments and deleting flags stay in the admin and the API.
Your data
What Claude reads from Slydog is sent to Anthropic under your own account with them, and their privacy policy covers it. See our privacy policy for what Slydog keeps when you connect.