@nexural/federation-server

Single MCP stdio server exposing the whole Nexural Federation knowledge base as one tool surface. Phase 11.x.

What it does

Wire ONCE into your editor's MCP config. Your agent gains two callable tools:

| Tool | Returns | | --------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------- | | federation_ask(query, kinds?, limit?) | Ranked excerpts from constitution, ADRs, warehouse docs, recipe THREAT_MODEL/DECISIONS/README, and eval sets — with source labels and citation paths | | federation_list_sources(kind?) | All sources currently indexed (for browsing or filtering subsequent asks) |

Setup

npm i -g @nexural/federation-server

In your editor's MCP config (Cursor / Claude Desktop / etc.):

{
  "mcpServers": {
    "nexural": {
      "command": "nexural-federation-server",
      "args": ["--root", "/Users/Sage/code/nexural/nexural-meta"]
    }
  }
}

Or set NEXURAL_META_ROOT in your env and omit --root:

{
  "mcpServers": {
    "nexural": {
      "command": "nexural-federation-server"
    }
  }
}

Usage from the agent's perspective

agent: I'll check the federation for what we've decided about cost discipline.
agent → tool: federation_ask({ query: "cost discipline LLM caps" })
tool → agent: {
  hits: [
    { source: "adr:0007-cost-guardrails-model-deprecation", title: "Cost Guardrails + Model Deprecation Handler", path: "docs/adr/0007-cost-guardrails-model-deprecation.md", snippet: "...streaming abort is mandatory..." },
    ...
  ]
}
agent: ADR-0007 has it covered. The pattern is X.

Compared to nx ask

nx ask is the CLI for humans. federation_ask (this server) is the same engine exposed to AI agents over MCP. Both query the same docs.

License

MIT.