action1-mcp
action1-mcp
MCP server for the Action1 RMM REST API. Gives Claude (or any other MCP client) read-only access to endpoints, missing updates, vulnerabilities, installed software, policies, automations and reports.
Unofficial — not affiliated with or endorsed by Action1 Corporation.
Why read-only
Action1's write surface deploys software and runs scripts on managed endpoints. That is remote code execution across a whole fleet, and it is not something to put behind a bearer token and a language model on day one. v1 exposes GETs only, and the guarantee is enforced in the HTTP client itself — not merely by the absence of write tools. See docs/ARCHITECTURE.md for the write tiers planned for v2 and SECURITY.md for the threat model.
Quick start
git clone https://github.com/molnkontakt/action1-mcp.git
cd action1-mcp
python3 -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"
# Credentials from the Action1 console: Configuration → API Credentials
export ACTION1_CLIENT_ID=...
export ACTION1_CLIENT_SECRET=...
export ACTION1_REGION=NorthAmerica # or Europe / Australia / NA-2
export ACTION1_DEFAULT_ORG_ID=... # optional, saves passing org_id everywhere
action1-mcp # stdio transportWire it into Claude Code:
claude mcp add action1 -- action1-mcpShared HTTP server
One long-lived process instead of one per client session:
export ACTION1_MCP_BEARER="$(openssl rand -base64 36 | tr -d '=+/' | cut -c1-48)"
action1-mcp --transport http --host 127.0.0.1 --port 3002{
"mcpServers": {
"action1": {
"type": "http",
"url": "https://mcp.example.com/action1/mcp",
"headers": { "Authorization": "Bearer <ACTION1_MCP_BEARER>" }
}
}
}Bind loopback and put a TLS-terminating reverse proxy in front. The server
refuses to bind a non-loopback address when ACTION1_MCP_BEARER is unset, but
that check is a backstop, not a deployment plan — see
docs/DEPLOY.md.
Configuration
Variable | Default | Purpose |
| — | Required. API credential from the Action1 console |
| — | Required. |
|
|
|
| unset | Org used when a tool is called without |
| unset | Bearer token required by the HTTP transport |
|
|
|
|
| HTTP bind address; |
|
| HTTP bind port; |
|
| Local throttle; Action1's ceiling is 30/min per tenant |
|
| Hard ceiling on items one tool call will page through. A tool's |
|
| Items per API request |
|
| Per-request timeout |
|
| Retries after a rate-limit response |
|
| Unblocks non-GET requests. No write tools exist in v1 |
Tools
21 read tools — see docs/TOOLS.md for the full reference.
Area | Tools |
Identity |
|
Tenant |
|
Endpoints |
|
Patching |
|
Vulnerabilities |
|
Automation |
|
Reports |
|
Every list tool returns the same envelope:
{"items": [...], "returned": 200, "total_items": 1432, "truncated": true}truncated: true means more data may exist beyond what came back. The list is a
prefix, not the whole set — narrow the query rather than concluding from a
partial answer. The flag errs towards true: several Action1 endpoints return no
total_items at all, and a missing counter is not evidence that the page you got
was everything.
Rate limiting
Action1 counts requests per enterprise, across every endpoint and every
integration hitting the tenant — so this server shares its budget with your
PowerShell scripts and any other tooling. It throttles locally below the
documented ceiling rather than discovering it through 429s, and when one arrives
anyway it honours the retry_after the API returns before falling back to
exponential backoff.
Development
pytest # no network: every request is served by an httpx MockTransport
ruff check .
mypy srcLicense
Apache-2.0. Provided as is; review before pointing it at a production tenant.