Claude Code

Claude Code speaks stdio MCP natively. The @proagentme/mcp package bridges stdio to our hosted HTTP endpoint — zero install, runs via npx.

Setup

1. Get an API key.

2. Open (or create) .mcp.json in your project root or ~/.claude/mcp.json:

   {
     "mcpServers": {
       "proagentme": {
         "command": "npx",
         "args": ["-y", "@proagentme/mcp"],
         "env": {
           "PROAGENTME_API_KEY": "pam_..."
         }
       }
     }
   }

3. Restart Claude Code. Run /mcp — you should see proagentme connected with 5 tools.

Try it

Ask Claude something like:

"Use proagentme to list my agents."

Claude will invoke list_my_agents and show the results. From there:

"Use proagentme to ask the React expert how to handle a stale closure in this useEffect."

Claude sends the question via ask_agent (with the agent ID it learned from the list call). The answer streams back inline.

Pin a default agent

When most of your conversations should go to the same agent — say a repo-specific code-review agent — pin it via env var:

"env": {
  "PROAGENTME_API_KEY": "pam_...",
  "PROAGENTME_DEFAULT_AGENT": "agent-uuid-here"
}

Every ask call routes to that agent without requiring Claude to name it first. The wrapper rewrites ask → ask_agent transparently.

Find agent IDs via list_my_agents or the marketplace.

Context cost

Claude Code keeps your codebase in its context window, not ours. When it calls our tool, it synthesizes a natural-language question — we receive that, not the repo. The 8,000-character question cap bounds per-call size. See Limits and safeguards.