Important
This repository has been retired. It is no longer maintained.
This package is now part of the AppliedIR/sift-mcp monorepo.
Documentation: appliedir.github.io/aiir
An MCP (Model Context Protocol) server providing comprehensive threat intelligence access to OpenCTI for Claude Code and other MCP clients.
Note: Validate and harden appropriately for your environment before production use.
This MCP is designed as a component of the Claude-IR AI-assisted incident response workstation.
git clone https://github.com/scriptedstatement/claude-ir.git
cd claude-ir
./setup.sh
claudeBenefits of Claude-IR installation:
- Guided setup with component selection
- Pre-configured MCP integration
- Works alongside forensic-rag-mcp (knowledge search) and windows-triage-mcp (file validation)
- Forensic discipline rules and investigation workflows
Note: This MCP requires an OpenCTI instance. See SETUP.md for guidance on connecting to or deploying OpenCTI.
Use standalone when you only need threat intelligence lookups without the full IR workstation.
git clone https://github.com/scriptedstatement/opencti-mcp.git
cd opencti-mcp
# Create virtual environment
python3 -m venv .venv
source .venv/bin/activate
# Install
pip install -e .
# Configure (requires OpenCTI instance - see SETUP.md)
export OPENCTI_TOKEN="your-api-token"
export OPENCTI_URL="http://localhost:8080" # Local Docker
# export OPENCTI_URL="https://opencti.example.com" # Remote/cloud
# Run server
python -m opencti_mcpFor OpenCTI setup guidance: See SETUP.md
| Category | Tools | Description |
|---|---|---|
| Unified Search | search_threat_intel |
Search across all entity types |
| Threats | search_threat_actor, search_campaign |
APT groups, campaigns |
| Arsenal | search_malware, search_tool, search_vulnerability |
Malware, tools, CVEs |
| Techniques | search_attack_pattern, search_course_of_action |
MITRE ATT&CK, mitigations |
| Observations | search_observable, search_sighting |
IOCs, detection events |
| Events | search_incident |
Security incidents |
| Analysis | search_reports, search_grouping, search_note |
Reports, groupings, notes |
| Entities | search_organization, search_sector |
Organizations, industries |
| Locations | search_location |
Countries, regions, cities |
| Infrastructure | search_infrastructure |
C2, hosting, botnets |
| Tool | Description |
|---|---|
lookup_ioc |
Get full IOC context with relationships |
lookup_hash |
Look up file hash (MD5/SHA1/SHA256) |
get_entity |
Get any entity by ID |
get_relationships |
Get entity relationships |
get_recent_indicators |
Get indicators from last N days |
| Tool | Description |
|---|---|
create_indicator |
Create new IOC |
create_note |
Add analyst note to entities |
create_sighting |
Record detection event |
trigger_enrichment |
Trigger VirusTotal/Shodan enrichment |
| Tool | Description |
|---|---|
get_health |
Check OpenCTI connectivity |
list_connectors |
List enrichment connectors |
get_network_status |
View adaptive metrics and recommendations |
force_reconnect |
Force reconnection (clears caches, resets circuit breaker) |
get_cache_stats |
View response cache statistics |
All search tools support advanced filtering:
{
"query": "APT29",
"limit": 10,
"offset": 0,
"labels": ["tlp:amber", "apt"],
"confidence_min": 70,
"created_after": "2024-01-01",
"created_before": "2024-12-31"
}Settings are loaded via Config.load() classmethod (config.py) with SecretStr token protection and helper parsers for typed env vars.
| Variable | Default | Description |
|---|---|---|
OPENCTI_URL |
http://localhost:8080 |
OpenCTI instance URL (use https:// for remote) |
OPENCTI_TOKEN |
- | API token (required) |
OPENCTI_READ_ONLY |
true |
Disable write operations |
OPENCTI_TIMEOUT |
60 |
Request timeout in seconds |
OPENCTI_MAX_RESULTS |
100 |
Maximum results per query |
OPENCTI_MAX_RETRIES |
3 |
Retry attempts for failures |
OPENCTI_RETRY_DELAY |
1.0 |
Initial retry delay (seconds) |
OPENCTI_RETRY_MAX_DELAY |
30.0 |
Maximum retry delay (seconds) |
OPENCTI_SSL_VERIFY |
true |
Verify SSL certificates (set false for self-signed) |
OPENCTI_CIRCUIT_THRESHOLD |
5 |
Failures before circuit opens |
OPENCTI_CIRCUIT_TIMEOUT |
60 |
Seconds before circuit recovery |
OPENCTI_EXTRA_OBSERVABLE_TYPES |
- | Custom observable types (comma-separated) |
OPENCTI_EXTRA_PATTERN_TYPES |
- | Custom pattern types (comma-separated) |
OPENCTI_LOG_FORMAT |
json |
Log format: "json" or "text" |
Control optional features via environment variables (prefix: FF_):
| Variable | Default | Description |
|---|---|---|
FF_STARTUP_VALIDATION |
true |
Test API connectivity on server start |
FF_RESPONSE_CACHING |
false |
Cache search results (reduces API calls) |
FF_GRACEFUL_DEGRADATION |
true |
Return cached results when service unavailable |
FF_NEGATIVE_CACHING |
true |
Cache "not found" results |
Option 1: Environment variable (recommended for production)
export OPENCTI_TOKEN="your-api-token"Option 2: Token file
mkdir -p ~/.config/opencti-mcp
echo "your-api-token" > ~/.config/opencti-mcp/token
chmod 600 ~/.config/opencti-mcp/tokenOption 3: .env file (development)
OPENCTI_TOKEN=your-api-token
If your OpenCTI instance has custom observable types or pattern types (e.g., proprietary IOC formats, additional detection languages), configure them via environment variables:
# Add custom observable types (case-sensitive, comma-separated)
export OPENCTI_EXTRA_OBSERVABLE_TYPES="Internal-Host,Cloud-Resource,Custom-IOC"
# Add custom pattern types (case-insensitive, comma-separated)
export OPENCTI_EXTRA_PATTERN_TYPES="osquery,kql,custom-sig"These extend the built-in allow-lists without removing standard STIX types.
Add to your project-local .mcp.json (or see the parent claude-ir project for automated setup):
{
"mcpServers": {
"opencti": {
"command": "/path/to/venv/bin/python",
"args": ["-m", "opencti_mcp"],
"cwd": "/path/to/opencti-mcp",
"env": {
"PYTHONPATH": "/path/to/opencti-mcp/src",
"OPENCTI_TOKEN": "your-api-token",
"OPENCTI_URL": "http://localhost:8080",
"OPENCTI_READ_ONLY": "true",
"OPENCTI_SSL_VERIFY": "true"
}
}
}
}opencti-mcp/
├── src/opencti_mcp/
│ ├── __init__.py # Package exports
│ ├── __main__.py # Entry point (with startup validation)
│ ├── server.py # MCP server (32 tools)
│ ├── client.py # OpenCTI API client (with caching)
│ ├── config.py # Configuration management
│ ├── validation.py # Input validation
│ ├── errors.py # Error hierarchy
│ ├── logging.py # Structured logging
│ ├── adaptive.py # Network metrics
│ ├── cache.py # TTL-based response caching
│ └── feature_flags.py # Feature flag management
├── tests/ # Test suite (1530 tests)
├── docs/ # Documentation
├── README.md # This file
├── CLAUDE.md # Development guide
├── IMPLEMENTATION.md # Technical architecture
└── pyproject.toml # Package configuration
# Install dev dependencies
pip install -e ".[dev]"
# Run all tests
pytest
# With coverage
pytest --cov=opencti_mcp --cov-report=html
# Type checking
mypy src/opencti_mcpnpx @anthropic/mcp-inspector python -m opencti_mcp# Run MCP server
python -m opencti_mcp
# Test connection (original CLI)
python opencti_query.py "APT29" --type threat_actor
# Quick health check
python -c "from opencti_mcp import OpenCTIClient, Config; c = OpenCTIClient(Config.load()); print('OK' if c.is_available() else 'FAIL')"The default OPENCTI_URL=http://localhost:8080 matches OpenCTI's standard Docker deployment, where the platform serves HTTP on port 8080. This is correct for local instances — traffic never leaves the machine.
For remote or cloud instances, use HTTPS. OpenCTI supports TLS either natively (APP__HTTPS_CERT__* env vars) or via a reverse proxy (Nginx, Caddy, Traefik) — the reverse proxy approach is more common in production.
export OPENCTI_URL=https://opencti.example.com # HTTPS for remote
export OPENCTI_TIMEOUT=120 # Higher for cloud (default 60 may be tight for complex queries)
export OPENCTI_MAX_RETRIES=3 # Retry on transient failures
export OPENCTI_SSL_VERIFY=true # Always for production (false only for self-signed certs)
export OPENCTI_READ_ONLY=true # Unless writes neededCloud users: If you experience timeouts or circuit breaker trips, increase
OPENCTI_TIMEOUTto 120-180. Complex threat intel queries on remote instances can take 60+ seconds under load.
Use get_network_status tool to view:
- Latency statistics (P50/P95/P99)
- Success rates
- Circuit breaker state
- Recommended timeout/retry settings
- Python 3.10+
- OpenCTI 6.x instance
- pycti 6.x
- mcp 1.x
Architecture and direction by Steve Anson. Implementation by Claude Code (Anthropic).
MIT