Per-repo documentation — each repo's docs/ and README, ingested and associated with the repo. Rendered through the <<<RepoDocs>>> tag.

mcp-coupler

A coupling device for any number of MCP servers — one MCP server that fronts N downstream MCP servers, merging all their tools into a single namespaced list and routing each call to the owning server. Register the coupler once; it fans out.

host ──stdio──► coupler ──stdio──► server A   (a1, a2 → A__a1, A__a2)
                        ──stdio──► server B   (b1     → B__b1)
                        ──stdio──► …any number…

Zero dependencies — MCP stdio is just newline-delimited JSON-RPC 2.0, implemented directly on Node's stdlib. Nothing to npm install.

Configure

$COUPLER_CONFIG (or ./coupler.config.json). The mcpServers key is also accepted, so a Claude Code config block drops in verbatim:

{
  "servers": {
    "congruency": { "command": "node", "args": ["/home/notificationsforsteven/congruency-mcp/server.js"] },
    "some-other": { "command": "npx", "args": ["-y", "@vendor/mcp"], "env": { "TOKEN": "…" } }
  }
}

Tool naming

Every downstream tool is exposed as <server>__<tool> (double-underscore), so two servers can each have a search tool without colliding. Descriptions are prefixed with [server]. Results pass through unchanged.

Built-in meta tools

- coupler__status — health + tool count of every downstream. - coupler__reconnect — tear down and re-spawn all downstreams, re-aggregate.

Behaviour

- Lazy + warm: downstreams are spawned on the first tools/list/tools/call (and warmed on initialize), then cached. - Resilient: a downstream that fails to start or dies is isolated and marked down in coupler__status; the others keep working. It never sinks the coupler. - Timeouts: 30 s handshake, 5 min per downstream tools/call. - Clean shutdown: on stdin close / SIGTERM it drains in-flight calls, kills all child servers, and exits.

Install into Claude Code

claude mcp add coupler -s local -- node /home/notificationsforsteven/mcp-coupler/coupler.js