InvGate Service Desk

invgate-service-desk-mcp

A Model Context Protocol server for InvGate Service Desk / Service Management.

Give your AI assistant full access to your InvGate Service Desk — query incidents, look up users, search the knowledge base, check assets, and manage tickets — all through natural language.

96 tools across 11 domains. Read-only by default, with optional write operations behind explicit opt-in.

Install in one click

The easiest way to use this server with Claude Desktop (and other MCPB-compatible clients) is the prebuilt bundle:

➡️ [Download `invgate-service-desk.mcpb`](https://github.com/tracegazer/invgate-service-desk-mcp/releases/latest/download/invgate-service-desk.mcpb) — then double-click it. Claude Desktop opens the installer, prompts for your InvGate base URL and API token, and you're done. No Python, no npx/uvx, no config files.

The .mcpb is attached to every GitHub Release and mirrored on Smithery. Prefer a package manager or container? See Quick start below.

What can it do?

63 read-only tools work out of the box. 33 write tools (incidents, KB, time tracking) activate only when you explicitly opt in.

Quick start

1. Install

pip install invgate-service-desk-mcp

Or run without installing (requires uv):

uvx invgate-service-desk-mcp

2. Connect to Claude Desktop

Add this to your claude_desktop_config.json:

{

"mcpServers": { "invgate": { "command": "uvx", "args": ["invgate-service-desk-mcp"], "env": { "INVGATE_BASE_URL": "https://acme.sd.cloud.invgate.net", "INVGATE_API_TOKEN": "your-api-token" } } } }

Restart Claude Desktop. That's it — start asking about your tickets.

Using pip install instead of uvx

{

"mcpServers": { "invgate": { "command": "invgate-service-desk-mcp", "env": { "INVGATE_BASE_URL": "https://acme.sd.cloud.invgate.net", "INVGATE_API_TOKEN": "your-api-token" } } } }

Enabling write operations

By default the server is read-only. Opt into writes with INVGATE_WRITE_PROFILE:

{

"mcpServers": { "invgate": { "command": "uvx", "args": ["invgate-service-desk-mcp"], "env": { "INVGATE_BASE_URL": "https://acme.sd.cloud.invgate.net", "INVGATE_API_TOKEN": "your-api-token", "INVGATE_WRITE_PROFILE": "support" } } } }

Compatibility: the legacy INVGATE_ENABLE_WRITES=1 still works and maps to full.

If both are set, the profile wins and a warning is printed to stderr. Note: support

deliberately keeps the Knowledge Base read-only. An invalid profile name fails fast at startup.

Warning: write mode lets the connected agent create, modify, and delete real content through your InvGate credential. There is no API to delete a ticket — created tickets can only be cancelled, not removed.

3. Get your API token

In your InvGate Service Desk instance: Settings > Integrations > API (or ask your admin). The server authenticates via HTTP Basic with username api and your token as the password.

Configuration

Configuration resolves in this order (highest priority first):

1. Environment variables (always win)
2. TOML config at ~/.config/invgate-service-desk-mcp/config.toml

# ~/.config/invgate-service-desk-mcp/config.toml

base_url = "https://acme.sd.cloud.invgate.net" api_token = "..."

api_username = "api"

write_profile = "none" # "none" (default) | "support" | "full"

telemetry_enabled = false

telemetry_detail = "metadata"

Tip: create the config directory first: mkdir -p ~/.config/invgate-service-desk-mcp

See config.toml.example for a copy-paste template.

Running the server

invgate-service-desk-mcp                 # STDIO transport (default)

invgate-service-desk-mcp --transport sse # SSE/HTTP transport

Security note: STDIO (the default) keeps everything local. The sse and streamable-http transports have no built-in authentication — only use them bound to loopback or behind an authenticated reverse proxy.

Observability (optional)

The server can emit OpenTelemetry traces, metrics, and logs — completely opt-in and vendor-neutral. Export to any OTLP-compatible backend (Dynatrace, Grafana, Datadog, Jaeger, etc.).

pip install "invgate-service-desk-mcp[telemetry]"

export INVGATE_TELEMETRY=1

OTLP endpoint and headers are configured via standard OpenTelemetry env vars (not in the TOML file):

Dynatrace setup

export OTEL_EXPORTER_OTLP_ENDPOINT="https://<your-env>.live.dynatrace.com/api/v2/otlp"

export OTEL_EXPORTER_OTLP_HEADERS="Authorization=Api-Token " export OTEL_EXPORTER_OTLP_METRICS_TEMPORALITY_PREFERENCE=delta export OTEL_SERVICE_NAME=invgate-service-desk-mcp

Token scopes needed: openTelemetryTrace.ingest, metrics.ingest, logs.ingest. See docs/observability-dynatrace.md for a detailed guide.

Generic OTLP collector

export OTEL_EXPORTER_OTLP_ENDPOINT="http://localhost:4318"

export OTEL_SERVICE_NAME=invgate-service-desk-mcp

Signals emitted:

• Traces — tool execution spans (GenAI semantic conventions) + InvGate API request spans with response size and item count
• Metrics — mcp.tool.duration, invgate.client.request.duration, mcp.tool.errors, invgate.response.item_count, invgate.response.size
• Logs — tool errors and unexpected API response shapes, correlated to traces (OTLP only, never stdout)

Development

git clone https://github.com/tracegazer/invgate-service-desk-mcp.git

cd invgate-service-desk-mcp uv venv && uv pip install -e ".[dev]" pytest

License

MIT