Deploy an MCP server
Put a remote MCP server online without rewriting working tool code into a platform specific agent framework. Tokay runs the existing Node or Python HTTP service as a long running process and gives its MCP endpoint an HTTPS hostname.
Last updated
AI agent? Start with llms.txt.
This is the MCP server your product owns. The official Tokay MCP is separate and lets your agent manage Tokay. The two servers never share tools or credentials.
Keep the server you already built
You can move from a local Streamable HTTP server to a remote endpoint with the same application code. Push or paste the source, and Tokay detects the runtime, web entry point, packages, environment variables, and supported databases used by the tools.
The official MCP SDK, FastMCP, and ordinary framework integrations run as normal application code. After deployment, the existing /mcp endpoint is ready to add to Claude, ChatGPT, Cursor, or another MCP client.
Long sessions run on a normal process
Tokay keeps streaming responses and stateful sessions attached to an always on server, so you do not have to adapt them to a short function invocation. The MCP Service runs on the Workspace's dedicated virtual machine and restarts if the process crashes.
When the tools expect Postgres, MySQL, MongoDB, or Redis, Tokay creates the resource in the Project and supplies its connection through the environment variable the code already reads.
Tokay stores tool credentials and authenticates MCP clients
You do not need to hardcode upstream credentials or build client authentication into every tool. Tokay stores API keys and database credentials as encrypted Project secrets and supplies them at runtime.
A backend or script that calls the server as itself uses a Machine credential scoped to the Project. An MCP client acting for a person uses Tokay Access OAuth, which a private app exposes automatically. The client discovers the /mcp resource, the person approves in the browser, and the token it receives works only for that Service and path. Both kinds of calls show up in the access log.
Neither credential can manage Tokay. The official Tokay MCP and the Platform API require their own separately authorized tokens.
Stdio servers need an HTTP transport
A server that only uses stdio is local by design because the client starts it as a subprocess. Add Streamable HTTP before hosting it remotely. The rest of the tool code can stay the same once it exposes an HTTP endpoint.
Your AI agent can deploy and test your server
Your coding agent can return a working MCP endpoint instead of leaving a code sample in the conversation. Through Tokay's GraphQL API, they can submit the source, set the write only credentials you provide, deploy the Service, read its URL, and connect with an MCP client to list and call the tools.
Point your agent to app.tokay.io/llms.txt for the deployment protocol.
Deploy your first app on Tokay's permanent free plan without a credit card.