@opvs-ai/mcp-runtime

@opvs-ai/mcp-runtime

Shared stdio MCP server used by OPVS per-skill MCP wrappers (@opvs-ai/mcp-agentboard, etc.). You usually don't import this directly — use a per-skill package instead.

API

import { runMcpServer } from "@opvs-ai/mcp-runtime";
import { tools, execute } from "@opvs-ai/skills-sdk/agentboard";

await runMcpServer({
  name: "opvs-agentboard",
  version: "0.1.0",
  tools,       // ToolSpec[] from @opvs-ai/skills-sdk/<skill>
  execute,     // dispatcher from the same skill
});

That's the entire surface. Transport is stdio JSON-RPC; auth flows through @opvs-ai/skills-sdk → @opvs-ai/core from ~/.opvs/config.json.

Writing a custom scoped MCP

If you want a different slice of OPVS tools than the three shipped per-skill packages, compose one yourself:

#!/usr/bin/env node
import { runMcpServer } from "@opvs-ai/mcp-runtime";
import { makeExecutor } from "@opvs-ai/skills-sdk";
import { tools as board } from "@opvs-ai/skills-sdk/agentboard";
import { tools as docs } from "@opvs-ai/skills-sdk/agentdocs";

const tools = [...board, ...docs];
await runMcpServer({ name: "opvs-board+docs", version: "0.1.0", tools, execute: makeExecutor(tools) });

This is how to stay under MCP client tool limits while including exactly the skills you need.

See also

• @opvs-ai/skills-sdk — where ToolSpec, executeTool, makeExecutor live
• @opvs-ai/mcp-agentboard, @opvs-ai/mcp-agentdocs, @opvs-ai/mcp-opvs-protocol — the three pre-built per-skill MCPs