Five minutes to your first call

Get a key, point your agent client at the MCP server, run one command to confirm it connected. Five clients below — pick yours.

Get an API key

Do this once. Everything below reuses the same exported key.

Sign in at aipocketagent.com and create a key under Settings → API. It starts with pa_live_. Export it once so the snippets below and your verify command both read the same value.

$ export POCKETAGENT_API_KEY="pa_live_…"

Claude Code

Claude Code reads MCP servers from a .mcp.json at the project root. Drop the server block in, paste your pa_live_ key, and the pocketagent tools are available the next time you start a session in that repo.

.mcp.json

{
  "mcpServers": {
    "pocketagent": {
      "command": "npx",
      "args": ["-y", "@pocketagent/mcp"],
      "env": { "POCKETAGENT_API_KEY": "pa_live_…" }
    }
  }
}

verify

$ claude mcp list

→ expect pocketagent: connected

Cursor

Cursor loads MCP servers from .cursor/mcp.json in the project (or ~/.cursor/mcp.json for every project). Add the block, then open Settings → MCP and confirm pocketagent shows a green dot.

.cursor/mcp.json

{
  "mcpServers": {
    "pocketagent": {
      "command": "npx",
      "args": ["-y", "@pocketagent/mcp"],
      "env": { "POCKETAGENT_API_KEY": "pa_live_…" }
    }
  }
}

verify

$ curl -sS -o /dev/null -w "%{http_code}\n" \
  https://aipocketagent.com/api/v1/brain \
  -H "Authorization: Bearer $POCKETAGENT_API_KEY"

→ expect 200

Continue

Continue configures MCP servers in ~/.continue/config.yaml under mcpServers. It runs in VS Code and JetBrains; reload the window after editing so it picks up the new server.

~/.continue/config.yaml

mcpServers:
  - name: pocketagent
    command: npx
    args: ["-y", "@pocketagent/mcp"]
    env:
      POCKETAGENT_API_KEY: pa_live_…

verify

$ curl -sS -o /dev/null -w "%{http_code}\n" \
  https://aipocketagent.com/api/v1/brain \
  -H "Authorization: Bearer $POCKETAGENT_API_KEY"

→ expect 200

OpenClaw

OpenClaw is a terminal agent in the Claude Code lineage; it reads MCP servers from ~/.openclaw/mcp.json. The block is identical to Claude Code's — same command, same env.

OpenClaw is young and the config path has moved between releases. If ~/.openclaw/mcp.json isn't read, check the MCP section of your openclaw config — the server block stays the same.

~/.openclaw/mcp.json

{
  "mcpServers": {
    "pocketagent": {
      "command": "npx",
      "args": ["-y", "@pocketagent/mcp"],
      "env": { "POCKETAGENT_API_KEY": "pa_live_…" }
    }
  }
}

verify

$ curl -sS -o /dev/null -w "%{http_code}\n" \
  https://aipocketagent.com/api/v1/brain \
  -H "Authorization: Bearer $POCKETAGENT_API_KEY"

→ expect 200

Hermes

Hermes loads MCP servers from ~/.config/hermes/mcp.json. Add the server, restart the agent, and the pocketagent tools register at startup alongside its built-ins.

Hermes is newer than the rest of this list. The config path may differ by version — point it at the MCP config its docs name; the server block is unchanged.

~/.config/hermes/mcp.json

{
  "mcpServers": {
    "pocketagent": {
      "command": "npx",
      "args": ["-y", "@pocketagent/mcp"],
      "env": { "POCKETAGENT_API_KEY": "pa_live_…" }
    }
  }
}

verify

$ curl -sS -o /dev/null -w "%{http_code}\n" \
  https://aipocketagent.com/api/v1/brain \
  -H "Authorization: Bearer $POCKETAGENT_API_KEY"

→ expect 200

Prefer to call the API directly instead of through a client? The API reference has every endpoint with a live try-it console. The MCP page covers the server itself, including the uvx install.