Every sign-up, exactly once
Deliveries are verified with the provider's signature. Duplicates, retries and replays never create a second contact.
Keyring is an open source CRM you host yourself. Every sign-up from your auth provider becomes a contact, enriched with job titles and company details, and ready for your team and your agents.
No CSV imports and no forms to fill in. Point your auth provider at Keyring once, and every person who signs up shows up as a contact a few seconds later.
Paste Keyring's webhook URL into your auth provider. Keyring checks every delivery's signature, and retries and replays land once.
Keyring looks each person and company up on context.dev: job title, location, logo, size and industry. Free-mail addresses are skipped, and fields a person edited are never overwritten.
Your team uses the web UI. Your agents use the MCP server or the REST API, which call the same code, so they see exactly what your team sees.
Settings shows each provider's webhook URL, the steps to connect it, and the last delivery or the reason it failed.
/webhooks/clerk/webhooks/workos/webhooks/better-auth/webhooks/supabase/webhooks/auth0/webhooks/stytchUsing something else? Post each sign-up from your backend to POST /v1/signups with an API key.
One container and a Postgres database. No per-seat pricing, no data leaving your infrastructure except the lookups you turn on.
Deliveries are verified with the provider's signature. Duplicates, retries and replays never create a second contact.
context.dev fills in people and companies. A daily cap keeps a leaked key or a flood of sign-ups from spending your credits.
An MCP server with 24 tools that agents sign in to with OAuth, and a REST API with read-only and expiring keys.
Sign in with a passkey, Google, GitHub or your own OIDC provider. Authenticator codes and recovery codes when you want them.
Migrations run on start, nothing is kept on disk, and several replicas can share one database. Run it on Unkey or any container host.
Webhook secrets and authenticator seeds are encrypted at rest, sign-in and webhook abuse is throttled per client, and every token is stored hashed.
Ask Claude who signed up from a company this week, or have a backend job tag every contact from a trial. Every change shows which agent made it, and for an MCP client, which person approved it.
Add the URL to Claude Code, Cursor or Claude Desktop. The client opens your browser, you approve it once, and it can search, read and update contacts as you.
# Claude Code claude mcp add --transport http keyring \ https://crm.example.com/mcp
Create a key per agent or backend. Make it read-only, give it an expiry, and let the agent revoke it itself if it ever leaks.
# Find everyone from a company curl "https://crm.example.com/v1/contacts?q=stripe.com" \ -H "Authorization: Bearer $KEYRING_API_KEY"
Unkey's MCP server lets a coding agent create the project, set the secrets, deploy and protect Keyring on Unkey, while you watch and approve. Sign in to Unkey from the agent, or create a free account on the way.
# Claude Code. Other agents: mcp.unkey.com/install claude mcp add --transport http unkey \ https://mcp.unkey.com
Deploy Keyring, an open source CRM (https://github.com/unkeyed/keyring), to Unkey for me, using Unkey's MCP server. The repository's docs/deploy-unkey.md is the reference for Keyring's settings.
0. Connect to Unkey
Check that Unkey's MCP tools are available (for example projects.listProjects). If they are not:
- Claude Code: run claude mcp add --transport http unkey https://mcp.unkey.com
then tell me to run /mcp, choose unkey, and Authenticate. MCP tools load when a session starts, so if they still do not appear, tell me to restart the session and paste this prompt again.
- Any other agent: tell me to follow https://mcp.unkey.com/install and stop until I say it is done.
Signing in opens Unkey in my browser, where I can also create an account. Do every Unkey step below through the MCP tools.
1. Ask me
- For a Postgres connection URL (Neon, Supabase, PlanetScale Postgres, RDS or similar), ideally in AWS us-east-1. If the provider gives a read/write role and a separate admin role, ask for both URLs.
- Whether I want enrichment, and if so for a context.dev API key.
2. Fork Keyring
Run gh repo fork unkeyed/keyring --clone and note the fork as owner/keyring.
3. Create the project and app on Unkey
- Create a project with slug keyring (projects.createProject).
- Create an app with slug default, sourced from my fork's default branch (apps.createApp with git.repository). If Unkey's GitHub App cannot see the fork, get the install link with github.installApp, give it to me, and retry once I have granted access.
- On the production environment (environments.updateSettings): Dockerfile "Dockerfile", root directory ".", port 8080, health check GET /healthz, and regions set to us-east-1 with 1 replica. The default region may be elsewhere, so set it explicitly.
4. Set production variables (environments.setEnvironmentVariables), every secret as writeonly
- DATABASE_URL: the read/write URL.
- ENCRYPTION_KEY: generate with openssl rand -base64 32. Save it to ~/.keyring-encryption-key with chmod 600 and tell me to copy it into my password manager. Unkey cannot show a writeonly value again, and Keyring cannot read its stored secrets without it.
- SETUP_TOKEN: generate with openssl rand -hex 32.
- CLIENT_IP_HEADER: X-Forwarded-For. Unkey's gateway sets it to the real client address, so Keyring throttles each client separately.
- CONTEXT_DEV_API_KEY: only if I gave one.
Do not set PORT. Never print a database password, and never write a secret into the repository.
5. Migrations
If I gave a separate admin URL, apply the migrations from the clone before the first deploy and keep the admin URL off Unkey:
MIGRATION_DATABASE_URL='the admin url' go run ./cmd/keyring migrate
Otherwise DATABASE_URL must be able to create tables; Keyring migrates when it starts.
6. Protect it at Unkey's gateway (gateway.setPolicies on production)
- Firewall, deny: request paths matching dotfiles and secrets (/.env, /.git, /.aws, /.ssh), backups and keys (.sql, .bak, .zip, .pem, .key), PHP and WordPress probes (.php, /wp-admin, /wp-login, /xmlrpc.php, /phpmyadmin, /cgi-bin) and appliance scanners (/actuator, /server-status, /HNAP1), case-insensitive. Never block /.well-known/, /v1/openapi.yaml, /setup/ or /invite/.
- Rate limits per remote IP: ^/oauth/(register|authorize)$ at 30 a minute, /auth/ at 120 a minute, /webhooks/ at 600 a minute.
7. Deploy
- Create a deployment on production (deployments.createDeployment) and follow it with deployments.getDeployment and deployments.listBuildLogs until it is ready.
- Set PUBLIC_URL to the stable address without a deployment id, https://keyring-.unkey.app, then deploy again. Variables and gateway policies only apply to a new deployment, and passkeys only work at PUBLIC_URL.
8. Check and hand over
- PUBLIC_URL/healthz and PUBLIC_URL/auth/setup/SETUP_TOKEN both answer 200, and PUBLIC_URL/.env answers 403.
- Give me PUBLIC_URL/setup/SETUP_TOKEN to create the first account with a passkey. Tell me to add a second passkey under Settings > Account and to connect my auth provider under Settings > Providers.
- Finish with a short summary: the URL, the Unkey project, the variables you set, where the encryption key is saved, and anything you could not do.