invgate-service-desk-mcp 0.2.0

InvGate Service Desk MCP — 96 tools across 11 domains (incidents, users, knowledge base, assets, time tracking), read-only by default with a 3-tier write access model (none/support/full), and built-in OpenTelemetry observability.

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 — 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?

Domain	Tools	Examples
Catalog	5	List priorities, statuses, incident types, categories (with search), sources
Incidents	34	Get ticket details, list by status/agent/customer, create & update tickets, reassign, comment, manage approvals
Users & Groups	7	Look up users, find by email/phone, list group members
Knowledge Base	10	Search articles, browse categories, create & update articles
Custom Fields	9	List field definitions, get options (list/tree), fields by category
Organization	11	Helpdesks, levels, locations, company structure
Assets / CIs	6	Find assets linked to incidents, CI relationships
Time Tracking	4	View logged hours, log new time entries
Triggers	2	List automation rules and their executions
Workflows	3	Inspect workflow fields, processes, and field values
Breaking News	5	View announcements, statuses, types

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:

Profile	Reads	Writes
none (default)	everything	nothing
support	everything	incidents (tickets, comments, reassign, approve) + time tracking
full	everything	incidents + time tracking + Knowledge Base

{
  "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

Env var	TOML key	Description
INVGATE_BASE_URL	base_url	Instance URL, e.g. https://acme.sd.cloud.invgate.net
INVGATE_API_TOKEN	api_token	API token (HTTP Basic password)
INVGATE_API_USERNAME	api_username	HTTP Basic username (optional, defaults to api)
INVGATE_WRITE_PROFILE	write_profile	Write access profile: none (default), support, or full
INVGATE_TELEMETRY	telemetry_enabled	Enable OpenTelemetry (default: false)
INVGATE_TELEMETRY_DETAIL	telemetry_detail	Span detail: metadata (default), ids, or full

# ~/.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 <YOUR_DT_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