Edit icon Edit This Page

Run a Model Context Protocol server over stdio for AI agents

Run a Model Context Protocol (MCP) server on stdin and stdout, so that an AI agent (Claude Code, Claude Desktop, Cursor, VS Code and other MCP clients) can use Repsy through rp. The client starts the process itself; you do not run it by hand. Configure the client to run:

rp mcp

and select the profile as for any other command (–hostname or REPSY_HOST). The server signs every request with the credential of that profile: REPSY_TOKEN when it is set, else the credential stored by rp auth login.

stdout carries the protocol only: rp prints nothing else there, never prompts and never reads anything but protocol messages from stdin. Messages for a person go to stderr. The server ends when the client closes stdin.

Three tools change the registry: create_repo, trigger_scan and create_deploy_token. They are offered only while the credential may use them: a login session (rp auth login) may use all three; an access token needs repo:manage for create_repo and create_deploy_token and repo:write for trigger_scan. A repo:read token sees only the read tools. The server checks the credential again while it runs and tells the client when the list changes. –read-only hides every tool that changes the registry, whatever the credential may do. A deploy token that create_deploy_token makes is in the tool’s result, so it enters the conversation and the transcript of the client.

Three more revoke, rotate or delete: revoke_deploy_token, rotate_deploy_token and delete_package_version (one version or Docker tag, never a whole package or repository). They follow the same rule and need repo:manage or a login session. Each repeats its target in a confirm argument, and clients always ask before calling one.

rp mcp tools lists the tools, and the reference page “MCP tools” describes each of them with its arguments.

Requests carry “mcp” in the User-Agent.

rp mcp [flags]

Options

      --read-only   Offer only the read tools, whatever the credential may do

Options inherited from parent commands

      --api-url string           Panel API URL of the profile, overrides the stored value
      --ca-file string           PEM file with extra trusted CA certificates
      --debug                    Print debug output with secrets redacted
      --hostname string          Repsy host to use instead of the current profile
      --insecure                 Do not verify TLS certificates (unsafe; plain http is chosen with the URL, not with this flag)
      --insecure-storage         Keep credentials in a plain 0600 file instead of the OS keyring (unsafe; a warning is printed once per command)
      --jq string                Filter JSON output with a jq expression (implies --output json)
      --no-color                 Disable colored output
      --no-input                 Never prompt; fail when input is needed
      --output table|json|yaml   Output format: table, json or yaml (default table; json when --jq is given)
      --owner string             Owner segment of Cloud protocol URLs; --owner= means none
      --registry-url string      Protocol host URL of the profile, overrides the stored value
      --type os|cloud            Product of the host, overrides detection
      --web-url string           Panel UI URL of the profile (where /cli/auth lives), overrides the stored value
      --yes                      Answer yes to confirmation prompts

See also

  • rp - Command line client for Repsy OS and Repsy Cloud
  • rp mcp tools - List the tools the MCP server offers
Found a problem with this page?