Genesys MCP

Vendor-neutral Model Context Protocol server for Genesys Cloud.

genesys-mcp is a Model Context Protocol (MCP) server that exposes Genesys Cloud platform capabilities — queues, conversations, users, presence, analytics — as MCP tools, resources, and prompts. Any MCP-compatible AI agent can speak to Genesys through it: Claude Desktop, Claude Code, Cursor, Continue, Cline, and any custom agent built on LangGraph, LangChain, CrewAI, AutoGen, or a raw provider SDK.

Status

v0.1.0 — pre-release. The framework, transports, and reliability layer are in place. The first batch of read-only tools lands in v0.2; curated writes (behind --enable-writes) and Resources/Prompts land in v0.3. The v1.0 launch ships everything together with PyPI, Docker (GHCR), an npx shim, and an MCPB bundle.

Why

The MCP ecosystem already has a Genesys Cloud server, but it ships stdio-only without HTTP transport, has no per-request auth context (a single module-global boolean gates the whole process), and has no production reliability layer (no 429 handling, no token refresh, no structured logging). genesys-mcp is designed for the production surface: hosted-ready Streamable HTTP, per-request auth, 429-aware retry, refresh-on-401, structured logs, and OpenTelemetry tracing — all from day one, all vendor-neutral.

Quickstart

Tools are not yet shipped — these snippets show the install pattern. v0.2 will list the actual tool surface here.

Claude Desktop

Add an entry under mcpServers in your claude_desktop_config.json:

{
  "mcpServers": {
    "genesys": {
      "command": "uvx",
      "args": ["genesys-mcp"],
      "env": {
        "GENESYS_REGION": "mypurecloud.com",
        "GENESYS_CLIENT_ID": "...",
        "GENESYS_CLIENT_SECRET": "..."
      }
    }
  }
}

Claude Code or Cursor

Add an entry to your editor's MCP server config (.claude/settings.json for Claude Code, the Cursor MCP settings panel for Cursor):

{
  "mcpServers": {
    "genesys": {
      "command": "genesys-mcp",
      "args": ["--transport", "stdio"],
      "env": {
        "GENESYS_REGION": "mypurecloud.com",
        "GENESYS_CLIENT_ID": "...",
        "GENESYS_CLIENT_SECRET": "..."
      }
    }
  }
}

Programmatic Python

Connect via the official mcp Python client over stdio or HTTP:

from mcp import ClientSession, StdioServerParameters
from mcp.client.stdio import stdio_client

params = StdioServerParameters(command="genesys-mcp", args=["--transport", "stdio"])

async with stdio_client(params) as (read, write):
    async with ClientSession(read, write) as session:
        await session.initialize()
        # tools = await session.list_tools()  # available from v0.2

For HTTP transport, run genesys-mcp --transport http --port 8000 and connect your MCP client to http://localhost:8000.

Install

pip install genesys-mcp

Coming v1.0: published on PyPI, GHCR, npm shim, and as an MCPB bundle attached to GitHub Releases.

Transports

Both transports share the same tool surface. Transport selection is environment-driven (MCP_TRANSPORT=stdio|http) or CLI-driven (--transport).

Security

The HTTP transport in v0.x ships unauthenticated. The MCP framing layer does not perform any caller authentication; the server trusts whatever process can reach the listening port. Operators have two responsible deployment options:

1. Bind to 127.0.0.1 only (the default MCP_HOST) and let local agents connect over loopback.
2. Bind to a non-loopback address only behind a reverse proxy (nginx, Caddy, Cloudflare Access, an API gateway) that performs authentication and authorisation before forwarding to genesys-mcp.

Binding to 0.0.0.0 without a fronting auth layer exposes the upstream Genesys credentials to anyone on the network. The Docker image therefore defaults MCP_HOST=127.0.0.1; operators who genuinely need a public bind must opt in explicitly with -e MCP_HOST=0.0.0.0.

A hosted, multi-tenant gateway with first-class authentication, audit, and RBAC is on the Pro/SaaS roadmap (see ADR-006 in docs/warm/decisions.md).

Trust model

A few things to know before installing plugins or running this in production:

• Entry-point plugins (the genesys_mcp.plugins group) execute in the server process with full access to every dependency, environment variable, and Genesys credential the server has. Only install plugins you trust.
• The license hook (genesys_mcp.license group) is a feature gate, not a security control. Any installed package can register one and any plugin can ignore the answer. It exists so the Pro package can express entitlements; treat it accordingly.
• OAuth client secrets are loaded from environment variables (and the .env file if present) and the server itself never logs them. We cannot guarantee the same of downstream plugins — vet them, or run with LOG_LEVEL=INFO and inspect output before trusting them with secrets.

Configuration

Configuration is read from environment variables (or a .env file). See .env.example for the full list. The most important variables:

Development

git clone https://github.com/digitalvanguardsolutions/genesys-mcp.git
cd genesys-mcp

uv sync --all-extras

uv run pytest
uv run ruff check src tests
uv run mypy src

uv run genesys-mcp --transport stdio

pre-commit install will wire up ruff, mypy, and a pre-push pytest run.

License

MIT — see LICENSE. Copyright 2026 Digital Vanguard Solutions.