Curated servers
| Server | Auth | What it does |
|---|---|---|
none | Query, inspect and explain your database from the agent. Paste a connection string: the gateway connects for you, nothing to install. | |
none | Read keys, scan, inspect hashes and lists, run allow-listed commands on your Redis or Valkey. Paste the URL: the gateway connects for you. | |
none | Run commands and read files on a server over SSH, with a command allow-list. The gateway holds the key; the agent never sees it. | |
none | Query, inspect and explain your MySQL or MariaDB. Paste a connection string: the gateway connects for you, nothing to install. | |
none | Query collections, inspect the schema and aggregate on your MongoDB. Paste the connection string: the gateway connects for you. | |
none | List buckets and objects, read and write objects on Amazon S3 or any S3-compatible store (MinIO, R2, Spaces). The gateway signs the requests for you. | |
none | Turn any HTTP API into agent tools. Give an OpenAPI URL and its operations become tools; without a spec, one request tool covers the base URL. | |
none | Run analytical SQL on your ClickHouse or ClickHouse Cloud. Paste the HTTPS URL with credentials: the gateway connects for you. | |
none | Search, count, inspect indices and mappings on your Elasticsearch or OpenSearch cluster. Paste the HTTPS URL with credentials or an API key. | |
none | List and inspect namespaces, pods, deployments and services, read pod logs, scale or restart. Paste a service-account token; the gateway calls the API server for you. | |
none | Query metrics, series and alerts on your Prometheus, Thanos, Mimir, VictoriaMetrics or Grafana Cloud endpoint. Read-only by nature. | |
none | Explore the schema and run queries on any GraphQL endpoint (Shopify, GitHub, Hasura, Apollo, your own). Mutations only when read-only is off. | |
none | Query and inspect your SQL Server or Azure SQL. Paste a connection string: the gateway connects for you, nothing to install. | |
API key (GITHUB_TOKEN) | Connect AI assistants to GitHub - manage repos, issues, PRs, and workflows through natural language. | |
OAuth | MCP server for Linear project management and issue tracking | |
OAuth | Official Notion MCP server | |
OAuth | Issues, events, releases. | |
OAuth | An MCP server for Vercel | |
OAuth | The Figma MCP server brings Figma design context directly into your AI workflow. | |
OAuth | Connect to Atlassian Jira, Confluence, and Compass to search, create, and manage your work. | |
OAuth | Workers, KV, D1, R2. | |
OAuth | MCP server integrating with Stripe - tools for customers, products, payments, and more. | |
OAuth | MCP server for interacting with the Supabase platform | |
API key (SLACK_BOT_TOKEN) | Channels, messages, search. | |
| Hugging Face | API key (HF_TOKEN) | Models, datasets, Spaces. |
OAuth | Sites, deploys, forms. | |
OAuth | 8,000+ apps through Zaps. | |
none | Up-to-date code docs for any prompt | |
your server | Browser automation on a Playwright MCP server you run over https. |
MCP Registry
16,152 remote servers published on the official MCP Registry (registry.modelcontextprotocol.io) are imported daily and searchable when you add a tool. Each one is probed regularly: the catalogue tells you whether it answered, and whether it wants a credential or OAuth. Adding a server only succeeds once it answered our probe with the credential you gave.
OAuth and provider apps
Most OAuth servers register Exorails automatically (dynamic client registration or our client metadata document), so authorizing is one click. A few providers, GitHub and Slack among them, only accept pre-registered apps: for those, paste a personal token or API key, or use your own OAuth app credentials.
Custom servers
- URL — https only. Private addresses, localhost and cloud metadata endpoints are refused, including after a redirect.
- Transport — Streamable HTTP (default) or legacy SSE.
- Authentication — none, a bearer token (
Authorization: Bearer …), a custom header (X-API-Key: …) or OAuth. Extra headers can reference secrets by name with{{secret:NAME}}. - Timeout — 120 seconds by default, adjustable per server.
Status
Every server is probed when added, then every 30 minutes, and whenever a call fails. ok means it answered tools/list; needs authorization means it rejected the credential; unreachable means it did not answer. The status is shown in the Tools tab and in tool errors the agent receives.