About ThreatLocker
ThreatLocker is an MCP server for teams using AI assistants to access the ThreatLocker Portal API. It covers computers, approval requests, audit logs, and organizations, with stateless requests and domain-based navigation.
Key features
- 01List computers and view details
- 02Get computer check-in history
- 03Filter approvals and count pending requests
- 04Search audit logs and file history
- 05List child organizations and retrieve organization data
- 06Navigate tools by domain
Use cases
- 01Review device details and check-in records
- 02Examine pending application approvals
- 03Investigate audit actions and file history
- 04Browse child organizations and prepare computer moves
MCP support
The MCP server exposes ThreatLocker Portal API operations for computers, approvals, audits, and organizations to MCP clients. It supports stdio and HTTP, including gateway mode with credentials passed in request headers.
Configuration
Local server config is the same for every client. Copy it into your client’s MCP config file.
docker run -i --rm -e THREATLOCKER_API_KEY -e THREATLOCKER_ORGANIZATION_ID -e MCP_TRANSPORT -e AUTH_MODE -e LOG_LEVEL ghcr.io/wyre-technology/threatlocker-mcp:v1.3.6THREATLOCKER_API_KEYSecretThreatLocker API keyTHREATLOCKER_ORGANIZATION_IDThreatLocker organization IDMCP_TRANSPORTTransport mode for the server. Set to 'stdio' for local CLI use; the image defaults to 'http' for gateway hosting.AUTH_MODESecretCredential source: 'env' reads vars locally, 'gateway' expects header injection from the WYRE MCP Gateway.LOG_LEVELLog verbosity: debug, info, warn, errorLocal servers need a runtime on your machine (e.g. Node.js 18+). Config file locations differ by client; see your client’s docs.
ghcr.io/wyre-technology/threatlocker-mcp:v1.3.6stdioTools
Local servers expose tools only once running on your machine, and the maintainer has not provided a tool list.
Details
From the project READMEThreatLocker MCP Server
A Model Context Protocol (MCP) server that provides AI assistants with access to the ThreatLocker Portal API. Manage computers, approval requests, audit logs, and organizations through natural language interactions.
Features
- Stateless Architecture: No session state required, fresh connections per request
- Decision-Tree Navigation: Navigate domains with
threatlocker_navigate - Gateway Mode: Multi-tenant support via HTTP headers
- Elicitation Support: Interactive prompts for missing parameters
- Comprehensive Error Handling: Detailed error messages and logging
- Docker Support: Production-ready containerization
Tools
Navigation
threatlocker_navigate- Navigate to a domain to see available toolsthreatlocker_status- Check API connection status and available domains
Computers
threatlocker_computers_list- List computers with filters (search, group, pagination)threatlocker_computers_get- Get detailed computer informationthreatlocker_computers_get_checkins- Get computer checkin history
Computer Groups
threatlocker_computer_groups_list- List computer groups with filtersthreatlocker_computer_groups_dropdown- Get computer groups for dropdown selection
Approval Requests
threatlocker_approvals_list- List approval requests with status filtersthreatlocker_approvals_get- Get detailed approval request informationthreatlocker_approvals_pending_count- Get count of pending approvalsthreatlocker_approvals_get_permit_application- Get permit application details
Audit Log
threatlocker_audit_search- Search audit log entries with filtersthreatlocker_audit_get- Get one audit log entry byeActionLogIdfrom a search row (deprecated alias:actionLogId)threatlocker_audit_file_history- Get audit history for a file. RequiresfullPathand eitherhostnameorcomputerId
Organizations
threatlocker_organizations_list_children- List child organizationsthreatlocker_organizations_get_auth_key- Get organization auth keythreatlocker_organizations_for_move_computers- Get organizations for computer moves
Configuration
Environment Variables
Stdio Mode (Direct API Access)
THREATLOCKER_API_KEY=your_api_key_here
THREATLOCKER_ORGANIZATION_ID=your_org_id_here
MCP_TRANSPORT=stdio
Gateway Mode (Multi-tenant)
AUTH_MODE=gateway
MCP_TRANSPORT=http
MCP_HTTP_PORT=8080
# Required. The HTTP server exits non-zero if this is empty.
CONDUIT_S2S_SECRET=gateway-provisioned-secret
# Optional. Unset binds 127.0.0.1. The container image sets 0.0.0.0.
# MCP_HTTP_HOST=127.0.0.1
CONDUIT_S2S_SECRET is required for the HTTP transport. The process logs an error and exits non-zero when it is unset. For local development only, MCP_ALLOW_INSECURE_DEV=1 starts the server anyway and logs a loud warning; do not set that flag in production. The secret is never written to the logs.
When MCP_HTTP_HOST is unset, the server binds 127.0.0.1. The Dockerfile sets MCP_HTTP_HOST=0.0.0.0 so the process listens inside the container; publish that port on localhost.
Gateway Mode Headers
When running in gateway mode, include these headers with each request:
X-Gateway-S2S: Service-to-service proof from the gateway (t=<unix>,v1=<hmac>). Requests without a valid header are rejected with 401 whenCONDUIT_S2S_SECRETis set.X-Threatlocker-Api-Key: Your ThreatLocker API key (required). A request without it is rejected with 401 and does not fall back toTHREATLOCKER_API_KEY.X-Threatlocker-Organization-Id: Your organization ID (optional; the API defaults to the key's primary organization)
/health and /healthz stay unauthenticated and do not read credentials. The stdio transport does not use CONDUIT_S2S_SECRET.
Logging
LOG_LEVEL=debug|info|warn|error # Default: info
Local Development
- Clone the repository:
git clone https://github.com/WYRE-AI/threatlocker-mcp.git
cd threatlocker-mcp
- Install dependencies:
npm install
- Set environment variables:
cp .env.example .env
# Edit .env with your ThreatLocker credentials
- Build and run:
npm run build
npm start
# Or for development with hot reload:
npm run dev
- Test the server:
# Stdio mode
echo '{"jsonrpc": "2.0", "id": 1, "method": "tools/list"}' | npm start
# HTTP mode. CONDUIT_S2S_SECRET is required unless you set
# MCP_ALLOW_INSECURE_DEV=1 (local development only; logs a loud warning).
# The server binds 127.0.0.1 unless MCP_HTTP_HOST is set.
MCP_TRANSPORT=http AUTH_MODE=gateway CONDUIT_S2S_SECRET=dev-secret npm run start:http
curl http://127.0.0.1:8080/health
Docker
Using Docker Compose
# Required. Compose will not start if this is unset or empty.
export CONDUIT_S2S_SECRET=gateway-provisioned-secret
# Pull and run latest image. The published port is 127.0.0.1:8080.
docker compose up -d
# Or build locally
docker compose -f docker-compose.dev.yml up --build
Using Docker directly
# Gateway mode (recommended). The container listens on 0.0.0.0 inside
# the network namespace; publish it on localhost only. The process
# exits if CONDUIT_S2S_SECRET is empty.
docker run -d \
--name threatlocker-mcp \
-p 127.0.0.1:8080:8080 \
-e AUTH_MODE=gateway \
-e CONDUIT_S2S_SECRET=gateway-provisioned-secret \
ghcr.io/wyre-ai/threatlocker-mcp:latest
# Stdio mode
docker run -d \
--name threatlocker-mcp \
-e THREATLOCKER_API_KEY=your_key \
-e THREATLOCKER_ORGANIZATION_ID=your_org_id \
-e MCP_TRANSPORT=stdio \
ghcr.io/wyre-ai/threatlocker-mcp:latest
Architecture
Directory Structure
src/
├── domains/ # Domain-specific handlers
│ ├── computers.ts
│ ├── computer_groups.ts
│ ├── approval_requests.ts
│ ├── audit_log.ts
│ ├── organizations.ts
│ ├── navigation.ts
│ └── index.ts
├── utils/ # Utilities
│ ├── client.ts # ThreatLocker API client
│ ├── logger.ts # Structured logging
│ ├── types.ts # TypeScript types
│ ├── server-ref.ts # Server reference for elicitation
│ └── elicitation.ts # Interactive prompts
├── server.ts # MCP server creation
├── index.ts # Stdio transport entry
└── http.ts # HTTP transport entry
Design Patterns
- Domain Handlers: Each API area has its own handler with
getTools()andhandleCall() - Lazy Loading: Domain handlers are imported on-demand
- Fresh Connections: New server instance per HTTP request for stateless operation
- Credential Invalidation: Client is reset when credentials change
- Elicitation Framework: Interactive prompts for missing parameters
License
Apache-2.0 - see LICENSE for details.
Comments
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.