RoboShadow MCP Server
# RoboShadow MCP Server
MCP server for [RoboShadow](https://www.roboshadow.com/)'s **Rubicon Platform Data API** - endpoint security (antivirus, Windows Defender, firewall), asset/hardware inventory, applications, users, vulnerabilities, and remediation status, for AI assistants and the WYRE Conduit gateway.
## Scope: every documented operation is read-only (deliberate, not a filtered subset)
RoboShadow's Rubicon Platform Data API (`https://cdn.development.roboshadow.com/publicapi/openapi.yaml`, fetched directly) defines **40 GET paths across 15 tags** (Antivirus, Applications, Defender, Devices, Disks, Firewall, Hardware, Identity, MFA, Remediation, Services, Updates, Users, Vulnerabilities, Vulnerability Scanner) and exactly **one** non-GET operation in the entire spec: `POST /identity/bearertoken`. That single POST is the vendor's own documented credential-refresh mechanism (OpenAPI tag: Authentication) - it exchanges a Refresh Token for a short-lived Bearer Token and mutates no organisation data. There is no create/update/delete/remediation-trigger endpoint anywhere in the published spec to exclude - unlike this session's other connectors, this one has no write surface to carve away from.
**This connector implements all 40 GET operations, one tool each, and calls `POST /identity/bearertoken` internally (never as an exposed tool) purely to keep its own access token fresh.** See Authentication below for why that call exists and why it isn't a tool.
## Authentication
RoboShadow's Rubicon Platform Data API authenticates with a **Bearer token** (`components.securitySchemes.BearerAuth`, `bearerFormat: JWT` in the vendor's own spec). Generate credentials at your RoboShadow portal under **Reports -> API Token**, which shows three values together: an **Organisation ID**, a **Bearer Token**, and a **Refresh Token**.
The Bearer Token itself is short-lived - RoboShadow's own `BearerTokenResponse` schema documents `expiresIn` in seconds (~1 hour in the vendor's example). A token pasted into a long-running connector would go stale within the hour, so **this connector stores the Refresh Token instead** and exchanges it for a fresh Bearer Token internally (`POST /identity/bearertoken`) before every data call, caching the result in-process until shortly before it expires. The Organisation ID is not stored as a credential - pass it as the `organisationId` argument to any tool (discover it with `roboshadow_list_organisations` if you don't already have it), since a Refresh Token can resolve to more than one organisation on an MSP/reseller account.
Two separate claims here, at deliberately different confidence levels - the two-tier split established across this batch of connectors:
- **Structurally verified (checked directly, stated with full confidence):** this connector's own code makes zero mutating calls against RoboShadow's data API. Every function in `client.ts` that reads organisation data calls one of the 40 dedicated GET operations in the published spec; the only non-GET call anywhere in `src/` is the bearer-token exchange itself, and it is never exposed as an MCP tool.
- **Vendor-documented, not independently verified (hedged deliberately):** RoboShadow's own OpenAPI spec defines no write/mutate endpoint for organisation data anywhere - there is no scoped, read-only variant of the Bearer/Refresh Token to request instead, because the full token grants exactly the same access the vendor's own spec grants: reads only. Whether some other, undocumented part of the RoboShadow platform (outside this Organisation-Endpoints spec) accepts write calls under the same token has not been tested by WYRE - only that this spec, and this connector, expose none.
In gateway mode the refresh token arrives per-request via the `X-RoboShadow-Refresh-Token` header and a missing header is rejected; the server does not fall back to `ROBOSHADOW_REFRESH_TOKEN`. In local/stdio mode that variable is read from the environment. HTTP mode also requires the `X-Gateway-S2S` signature described below.
## Configuration
| Env var | Description |
|---|---|
| `ROBOSHADOW_REFRESH_TOKEN` | RoboShadow Refresh Token (Reports -> API Token in the portal). Required for stdio and for `AUTH_MODE=env`. Not read in gateway mode. |
| `MCP_TRANSPORT` | `stdio` (default) or `http`. |
| `AUTH_MODE` | `env` (default, reads the var above) or `gateway` (credential arrives per-request via the `X-RoboShadow-Refresh-Token` header, injected by the Conduit gateway). Gateway mode returns 401 when that header is missing and never falls back to `ROBOSHADOW_REFRESH_TOKEN`. The container image and Compose default to `gateway`. |
| `CONDUIT_S2S_SECRET` | Required to start the HTTP transport. HMAC secret checked against `X-Gateway-S2S` on every `/mcp` request. If empty or a known placeholder (`changeme`, `your-s2s-secret`, and the other public examples), the process logs an error and exits non-zero. The value is never logged. Compose will not start when it is unset. |
| `MCP_ALLOW_INSECURE_DEV` | Set to exactly `1` only for local development, to start HTTP without `CONDUIT_S2S_SECRET`. The process logs a warning, does not check `X-Gateway-S2S`, and binds `127.0.0.1` only. A wider `MCP_HTTP_HOST` (including `0.0.0.0`) is ignored. A known placeholder secret is still refused. |
| `MCP_HTTP_HOST` | Bind address when `MCP_TRANSPORT=http`. Defaults to `127.0.0.1`. Honored only when `CONDUIT_S2S_SECRET` is set. The container image sets `0.0.0.0`. |
| `MCP_HTTP_PORT` | HTTP port. Defaults to `8080`. |
| `LOG_LEVEL` | `debug` \| `info` (default) \| `warn` \| `error`. |
`/health` stays unauthenticated and does not read credentials. The stdio transport does not use `CONDUIT_S2S_SECRET`.
### X-Gateway-S2S
With `MCP_TRANSPORT=http` and `CONDUIT_S2S_SECRET` set, every `/mcp` request must include a gateway-signed `X-Gateway-S2S` header:
```
X-Gateway-S2S: t=<unix seconds>,v1=<hex hmac-sha256>
```
`v1` is HMAC-SHA256 of the literal string `t=<unix seconds>` (the same `t=` prefix and value that appear in the header), keyed with `CONDUIT_S2S_SECRET`, and encoded as lowercase hex. The timestamp must be within 300 seconds of the server clock. A missing or invalid header returns 401 and the handler does not read vendor credentials.
In `AUTH_MODE=gateway`, `/mcp` also requires `X-RoboShadow-Refresh-Token`. A missing or empty header returns 401. The server does not fall back to `ROBOSHADOW_REFRESH_TOKEN` in the process environment.
### Request limits
- HTTP request bodies are capped at 256 KiB (262144 bytes). After the `X-Gateway-S2S` check passes, a `Content-Length` above the cap returns 413 with a JSON-RPC error; bodies without `Content-Length` are capped by the MCP SDK transport. Unauthenticated requests still get 401 first.
- `tools/call` arguments are capped at 10,000 elements (every array item and object key at any depth). Larger inputs return an `isError` tool result and are never dispatched.
These limits address CVE-2026-111768 / GHSA-46jr-4wx8-v453 (memory exhaustion during validation) and GHSA-f897-62hf-6gxq (no request body / batch cap).
## Tools
40 read-only tools, one per GET operation in the Rubicon Platform Data API's published spec. Every tool that reads organisation-scoped data takes `organisationId` - discover it with `roboshadow_list_organisations`.
### Identity
- `roboshadow_list_organisations` - list every organisation visible to this refresh token.
- `roboshadow_list_organisation_users` - list an organisation's users (display name, email, permission level). **PII - admin-gated.**
### MFA
- `roboshadow_get_mfa_report` - paged MFA status report (registered methods, MFA status, Global Administrator flag) for a Microsoft-identity-linked organisation. **PII/identity data - admin-gated.**
### Vulnerability Scanner
- `roboshadow_list_scan_status` - status of every external vulnerability scan (VScanner), filterable by schedule/status.
- `roboshadow_get_scan_summary` - detailed summary of one scan (IP + web vulnerability totals).
- `roboshadow_get_scan_ips_overview` - per-target CVE/port breakdown for one scan.
### Applications
- `roboshadow_list_application_groups` - applications grouped across the organisation, with install/device/version counts.
- `roboshadow_get_application_group` - detail for one application group.
- `roboshadow_get_device_applications` - applications installed on one device.
### Devices
- `roboshadow_list_devices` - every device in the organisation, with hardware/OS identity, network address, health scores, and CVE/update counts. **PII (IP/MAC address, logged-on user, serial number) - admin-gated.**
- `roboshadow_get_device` - detail for one device. **PII - admin-gated.**
- `roboshadow_get_device_user_profiles` - Windows user profiles on one device (includes SID). **PII - admin-gated.**
- `roboshadow_get_device_users` - local/AD users on one device (email, employee ID, admin flag). **PII - admin-gated.**
### Antivirus / Defender
- `roboshadow_get_device_antivirus` - antivirus/endpoint-protection state for one device.
- `roboshadow_get_antivirus_report_by_device` - antivirus status for every device.
- `roboshadow_get_ransomware_report` - ransomware-protection status for every device.
- `roboshadow_get_antivirus_summary` - organisation-wide antivirus summary.
- `roboshadow_get_antivirus_threats` - detected antivirus threats across the organisation.
- `roboshadow_get_windows_defender_report` - Windows Defender configuration status for every device.
### Firewall
- `roboshadow_get_firewall_report` - Windows firewall status (per profile) for every device.
### Disks
- `roboshadow_get_device_disks` - disks on one device.
- `roboshadow_get_disk_shares` - network shares exposed on one disk.
- `roboshadow_get_disk_report` - disk usage across the organisation, grouped by disk.
- `roboshadow_get_disk_report_by_device` - disk information grouped by device.
- `roboshadow_get_disk_summary` - organisation-wide disk-usage summary (low-space devices, unencrypted drives, shares, USB disks).
### Hardware
- `roboshadow_get_hardware_report_by_device` - hardware inventory grouped by device.
- `roboshadow_get_hardware_summary` - organisation-wide hardware summary.
### Services
- `roboshadow_get_device_services` - Windows services on one device, including unquoted-service-path vulnerability flag.
### Updates
- `roboshadow_get_device_updates` - missing Windows updates for one device.
- `roboshadow_get_updates_report_by_device` - paginated missing-updates report grouped by device.
- `roboshadow_get_updates_summary` - organisation-wide update summary and top devices needing attention.
### Vulnerabilities
- `roboshadow_get_device_vulnerability_summary` - per-application CVE detail and severity rollup for one device.
- `roboshadow_get_vulnerable_applications` - applications with known vulnerabilities across the organisation.
- `roboshadow_get_cpe_vulnerabilities` - CPE vulnerability findings across the organisation.
- `roboshadow_get_cve_vulnerabilities` - CVEs affecting the organisation (CVSS/EPSS, known-exploited flag).
- `roboshadow_get_cve_fixes` - available fixes/patches for one CVE.
- `roboshadow_get_vulnerable_devices` - devices with known vulnerabilities.
- `roboshadow_get_vulnerabilities_summary` - organisation-wide vulnerability summary and top-5 critical apps/CVEs/devices.
### Remediation
- `roboshadow_get_remediation_report` - vulnerability-remediation attempts across the organisation (action, target, outcome).
- `roboshadow_get_remediation_attempt_cves` - CVEs addressed by one remediation attempt.
## Sensitivity
RoboShadow is an endpoint-security and vulnerability-management platform - even as plain reads, most of this connector's 40 tools surface data that is PII (device users' SIDs, employee IDs, and email addresses; a device's logged-on user, IP/MAC address, and serial number), identity/MFA posture, or security-assessment data (vulnerabilities, threats, remediation outcomes, external scan results). Per this session's convention for vendors of this class (the same posture as Cork, CyberQP, Cisco Duo, and Slide in this same wave), **every tool in this connector is `isAdmin: true` in the Conduit wiring, regardless of verb** - this is a deliberate, vendor-wide design decision documented once in `vendor-config.ts`'s `roboshadow` entry, not a per-tool sensitive-read escalation.
## Development
```bash
npm install
npm run build
npm test
npm run lint # tsc --noEmit
```
## Docker
The image listens on `0.0.0.0` inside the container (`MCP_HTTP_HOST`) and defaults to `AUTH_MODE=gateway`. Startup still refuses to run unless `CONDUIT_S2S_SECRET` is a real secret. Publish the port on loopback. In gateway mode the refresh token arrives on each request as `X-RoboShadow-Refresh-Token`; it is not read from the environment.
```bash
docker build -t roboshadow-mcp .
docker run \
-e CONDUIT_S2S_SECRET="$CONDUIT_S2S_SECRET" \
-p 127.0.0.1:8080:8080 \
roboshadow-mcp
```
Compose refuses to start when `CONDUIT_S2S_SECRET` is unset, defaults `AUTH_MODE` to `gateway`, publishes `127.0.0.1:8080:8080`, and does not set `ROBOSHADOW_REFRESH_TOKEN`:
```bash
CONDUIT_S2S_SECRET="$CONDUIT_S2S_SECRET" docker compose up
```
Single-tenant HTTP (`AUTH_MODE=env`) still requires the S2S secret, then uses `ROBOSHADOW_REFRESH_TOKEN`:
```bash
docker run \
-e CONDUIT_S2S_SECRET="$CONDUIT_S2S_SECRET" \
-e AUTH_MODE=env \
-e ROBOSHADOW_REFRESH_TOKEN="$ROBOSHADOW_REFRESH_TOKEN" \
-p 127.0.0.1:8080:8080 \
roboshadow-mcp
```
Local HTTP without a secret is development-only. It binds `127.0.0.1` even if `MCP_HTTP_HOST` is `0.0.0.0`, and logs a warning:
```bash
MCP_TRANSPORT=http MCP_ALLOW_INSECURE_DEV=1 npm run start:http
```
## License
Apache-2.0
TDQS
Scored across 40 tools
Most tools target a distinct resource or aggregation level, and the descriptions are precise about scope. However, the many report/summary/by-device variants for antivirus, disk, updates, and vulnerabilities create moderate risk of an agent selecting the wrong one.
Every tool follows the same roboshadow_verb_noun pattern with lowercase snake_case, using get for detail and list for collection-style operations. The naming is highly predictable across all 40 tools.
40 tools is well above the 25+ threshold and feels heavy for the domain. Many tools are report variants that could be consolidated into parameterised endpoints, though each does map to a distinct data view.
The tool surface provides broad read-only coverage of organisations, devices, vulnerabilities, updates, disks, and remediation history. It lacks write/action capabilities such as initiating scans or applying fixes, but that appears consistent with a read-only security reporting server.