@vertiman/marky

Stdio bridge to the Marky MCP HTTP endpoint. Forwards JSON-RPC from stdin to Marky's HTTPS /mcp endpoint and writes the response to stdout, so any MCP client that speaks stdio can talk to Marky.

Use with Claude Code

claude mcp add marky -- npx -y @vertiman/marky \
  --env MARKY_ENDPOINT=https://mcpdocs.getlegio.com/mcp \
  --env MARKY_TOKEN=mk_live_...your_token...

Mint a token at https://mcpdocs.getlegio.com/settings.

Environment

| Var | Default | Description | |---|---|---| | MARKY_ENDPOINT | https://mcpdocs.getlegio.com/mcp | HTTP MCP endpoint URL. | | MARKY_TOKEN | — (required) | Bearer token, e.g. mk_live_.... |

Project detection

On each request the shim detects a project slug from the current working directory (basename of git remote get-url origin, or basename $PWD as a fallback) and sends it on the x-marky-project header. This lets you pin the active project per repo without re-configuring the MCP server.

How it works

Reads JSON-RPC messages from stdin → HTTPS POST to MARKY_ENDPOINT with Authorization: Bearer $MARKY_TOKEN → response echoed to stdout. No state between requests; the shim is a pure forwarder.

Input framing is auto-detected on the first byte: clients that send LSP-style Content-Length: headers (e.g. Codex) are handled transparently, and clients that send newline-delimited JSON (the MCP spec stdio default, e.g. Claude Code) work too. Output mirrors whichever framing the client used.