Action1

Action1 MCP Server

MCP server for Action1 — endpoint inventory, patch visibility, and policy automation via the Model Context Protocol.

Read-only in v1. Deploy / automation surface is intentionally separated to a later release for blast-radius reasons (a bad policy push can brick endpoint fleets).

Tools

API surface maps to PSAction1 (Action1's MIT-licensed PowerShell module). When the v1 surface earns its keep, write tools (deploy, requery, package upload) come in v2 behind separate review.

Interactive Device Card (MCP Apps)

action1_get_endpoint renders as an interactive device-summary card in MCP Apps hosts (Claude Desktop/web) — status, OS, user, IP, agent version, reboot-required, and missing-update counts at a glance; plain-JSON behavior is unchanged in other hosts. The card is read-only (matching the v1 tool surface), neutral by default, and brandable via window.__BRAND__ injection or MCP_BRAND_* env vars (MCP_BRAND_NAME, MCP_BRAND_LOGO_URL, MCP_BRAND_PRIMARY_COLOR, MCP_BRAND_ACCENT_COLOR, MCP_BRAND_BG, MCP_BRAND_TEXT) — no rebuild needed.

Usage

Claude Desktop (MCPB)

Install via the MCPB bundle from the latest release.

Required credentials (created in Action1 → Settings → API Credentials, non-recoverable on creation — copy immediately):

• API Key (Client ID)
• Secret
• Region (NorthAmerica default; also NorthAmerica-2, Europe, Australia)
• Optional default organization id (for single-tenant use)

Stdio (direct)

ACTION1_API_KEY=... \

ACTION1_SECRET=... \ ACTION1_REGION=NorthAmerica \ ACTION1_DEFAULT_ORG_ID=org-... \ npx -y github:WYRE-AI/action1-mcp

HTTP

CONDUIT_S2S_SECRET is required. If it is empty or a known placeholder (changeme, replace-with-a-real-secret, and similar), the HTTP server logs an error and exits non-zero. It never prints the secret. Generate one with openssl rand -hex 32.

MCP_HTTP_HOST defaults to 127.0.0.1 when unset. Set 0.0.0.0 only when CONDUIT_S2S_SECRET is set. The container image does that and sets AUTH_MODE=gateway. MCP_HTTP_PORT (or PORT) selects the port and defaults to 8080.

MCP_ALLOW_INSECURE_DEV=1 starts HTTP without CONDUIT_S2S_SECRET, logs a warning, skips X-Gateway-S2S, and binds 127.0.0.1 only. A non-loopback MCP_HTTP_HOST (including 0.0.0.0) refuses to start. A placeholder secret is still rejected. Local development only.

/health stays unauthenticated and does not read credentials. The stdio transport does not use CONDUIT_S2S_SECRET.

X-Gateway-S2S

Every HTTP request except /health must include X-Gateway-S2S when a secret is configured. The value is t= ,v1= , where the hex is HMAC-SHA256 of the literal string t= keyed with CONDUIT_S2S_SECRET (64 lowercase hex characters). The timestamp must be within 300 seconds of the server clock. A missing or invalid header is HTTP 401, before vendor credentials are read.

TS=$(date +%s)

SIG=$(printf 't=%s' "$TS" | openssl dgst -sha256 -hmac "$CONDUIT_S2S_SECRET" -hex | awk '{print $NF}') curl -sS http://127.0.0.1:8080/ \ -H 'Content-Type: application/json' \ -H 'Accept: application/json, text/event-stream' \ -H "X-Gateway-S2S: t=${TS},v1=${SIG}" \ -H "X-Action1-API-Key: $ACTION1_API_KEY" \ -H "X-Action1-Secret: $ACTION1_SECRET" \ -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Gateway mode (AUTH_MODE=gateway)

The Docker image and Compose file default to gateway mode. Every request must include:

• X-Action1-API-Key
• X-Action1-Secret
• X-Action1-Region (optional)
• X-Action1-Default-Org-Id (optional)

A request missing the API key or secret gets 401 and does not fall back to ACTION1_API_KEY / ACTION1_SECRET in the process environment. Those headers are the only credential source for that request.

docker run \

-e CONDUIT_S2S_SECRET="$CONDUIT_S2S_SECRET" \ -p 127.0.0.1:8080:8080 \ ghcr.io/wyre-ai/action1-mcp:latest

Compose refuses to start when CONDUIT_S2S_SECRET is unset, publishes 127.0.0.1:8080:8080, and does not pass static Action1 credentials into the container:

CONDUIT_S2S_SECRET="$CONDUIT_S2S_SECRET" docker compose up

Single-tenant HTTP (AUTH_MODE=env)

Requests still require a valid X-Gateway-S2S header. Vendor credentials come from ACTION1_API_KEY and ACTION1_SECRET when the request does not carry those headers.

docker run \

-e CONDUIT_S2S_SECRET="$CONDUIT_S2S_SECRET" \ -e AUTH_MODE=env \ -e ACTION1_API_KEY="$ACTION1_API_KEY" \ -e ACTION1_SECRET="$ACTION1_SECRET" \ -e ACTION1_REGION=NorthAmerica \ -p 127.0.0.1:8080:8080 \ ghcr.io/wyre-ai/action1-mcp:latest

Local HTTP without a secret is development-only. It binds 127.0.0.1 and logs a warning:

MCP_TRANSPORT=http MCP_ALLOW_INSECURE_DEV=1 npm start

Architecture

src/

├── index.ts # stdio + HTTP transports, tool dispatch, S2S startup ├── s2s-verify.ts # X-Gateway-S2S HMAC-SHA256 check ├── card.builder.ts # MCP Apps device-card normalization + brand injection ├── resources.ts # resources/list + resources/read (ui:// device card) ├── generated/ │ └── device-card-html.ts # committed vite single-file bundle (npm run build:ui) ├── sdk/ │ └── action1-client.ts # embedded REST + OAuth client (factor-out candidate │ # if surface crosses ~20 tools / 2+ domains) ├── utils/ │ ├── client.ts # credential resolution (env vs gateway headers) │ └── types.ts # DomainHandler interface ├── domains/ # one file per resource type │ ├── organizations.ts │ ├── endpoints.ts │ ├── policies.ts │ └── updates.ts └── tests/ # domain tests + MCP Apps contract tests ui/ # device-card source (index.html + device-card.ts)

Per-request credential isolation via AsyncLocalStorage — concurrent requests in HTTP mode never share credentials through process.env.

Development

npm install

npm run build npm test npm run dev # tsc --watch npm run lint # eslint npm run typecheck

License

Apache-2.0. See LICENSE.