GHSA-73HC-M4HX-79PJ
Vulnerability from github – Published: 2026-03-05 21:42 – Updated: 2026-03-09 15:50Summary
The /api/health/detailed endpoint returns detailed system information including OS version, Python version, CPU count, memory totals, disk usage, and the full database filesystem path. When MCP_ALLOW_ANONYMOUS_ACCESS=true is set (required for the HTTP server to function without OAuth/API key), this endpoint is accessible without authentication. Combined with the default 0.0.0.0 binding, this exposes sensitive reconnaissance data to the entire network.
Details
Vulnerable Code
health.py:90-101 - System information collection
system_info = {
"platform": platform.system(), # e.g., "Linux", "Darwin"
"platform_version": platform.version(), # Full OS kernel version string
"python_version": platform.python_version(),# e.g., "3.12.1"
"cpu_count": psutil.cpu_count(), # CPU core count
"memory_total_gb": round(memory_info.total / (1024**3), 2),
"memory_available_gb": round(memory_info.available / (1024**3), 2),
"memory_percent": memory_info.percent,
"disk_total_gb": round(disk_info.total / (1024**3), 2),
"disk_free_gb": round(disk_info.free / (1024**3), 2),
"disk_percent": round((disk_info.used / disk_info.total) * 100, 2)
}
health.py:131-132 - Database path disclosure
if hasattr(storage, 'db_path'):
storage_info["database_path"] = storage.db_path # Full filesystem path
Authentication Bypass Path
The /api/health/detailed endpoint uses require_read_access which calls get_current_user. When MCP_ALLOW_ANONYMOUS_ACCESS=true, the auth middleware grants access:
# middleware.py:372-379
if ALLOW_ANONYMOUS_ACCESS:
logger.debug("Anonymous access explicitly enabled, granting read-only access")
return AuthenticationResult(
authenticated=True,
client_id="anonymous",
scope="read",
auth_method="none"
)
Note: The basic /health endpoint (line 68) has no auth dependency at all and returns version and uptime information unconditionally.
Information Exposed
| Field | Example Value | Reconnaissance Value |
|---|---|---|
platform |
"Linux" |
OS fingerprinting |
platform_version |
"#1 SMP PREEMPT_DYNAMIC..." |
Kernel version → CVE targeting |
python_version |
"3.12.1" |
Python CVE targeting |
cpu_count |
8 |
Resource enumeration |
memory_total_gb |
32.0 |
Infrastructure profiling |
database_path |
"/home/user/.mcp-memory/memories.db" |
Username + file path disclosure |
database_size_mb |
45.2 |
Data volume estimation |
Attack Scenario
- Attacker scans the local network for services on port 8000
- Finds mcp-memory-service with HTTP enabled and anonymous access
- Calls
GET /api/health/detailed(no credentials needed) - Receives OS version, Python version, full database path (revealing username), system resources
- Uses this information to:
- Target known CVEs for the specific OS/Python version
- Identify the database file location for potential direct access
- Profile the system for further attacks
PoC
# Show the system info that would be exposed
import platform, psutil
system_info = {
"platform": platform.system(),
"platform_version": platform.version(),
"python_version": platform.python_version(),
"cpu_count": psutil.cpu_count(),
"memory_total_gb": round(psutil.virtual_memory().total / (1024**3), 2),
}
print(system_info) # All of this is returned to unauthenticated users
Impact
- OS fingerprinting: Exact OS and kernel version enables targeted exploit selection
- Path disclosure: Database path reveals username, home directory structure, and file locations
- Resource enumeration: CPU, memory, and disk info reveal infrastructure scale
- Reconnaissance enablement: Combined information significantly reduces attacker effort for follow-up attacks
Remediation
- Remove system details from default health endpoint - return only
status,version,uptime:
@router.get("/health/detailed")
async def detailed_health_check(
storage: MemoryStorage = Depends(get_storage),
user: AuthenticationResult = Depends(require_write_access) # Require admin/write access
):
# Only return storage stats, not system info
...
- Do not expose
database_path- this leaks the filesystem structure:
# Remove or redact
# storage_info["database_path"] = storage.db_path # REMOVE THIS
- Add auth to basic
/healthor limit it to status-only (no version):
@router.get("/health")
async def health_check():
return {"status": "healthy"} # No version, no uptime
Alternatively, Bind to 127.0.0.1 by default instead of 0.0.0.0, preventing network-based reconnaissance entirely:
# In config.py — change default from '0.0.0.0' to '127.0.0.1'
HTTP_HOST = os.getenv('MCP_HTTP_HOST', '127.0.0.1')
Users who need network access can explicitly set MCP_HTTP_HOST=0.0.0.0, making the exposure a conscious opt-in rather than a default.
{
"affected": [
{
"package": {
"ecosystem": "PyPI",
"name": "mcp-memory-service"
},
"ranges": [
{
"events": [
{
"introduced": "0"
},
{
"fixed": "10.21.0"
}
],
"type": "ECOSYSTEM"
}
]
}
],
"aliases": [
"CVE-2026-29787"
],
"database_specific": {
"cwe_ids": [
"CWE-200"
],
"github_reviewed": true,
"github_reviewed_at": "2026-03-05T21:42:10Z",
"nvd_published_at": "2026-03-07T16:15:55Z",
"severity": "MODERATE"
},
"details": "### Summary\nThe `/api/health/detailed` endpoint returns detailed system information including OS version, Python version, CPU count, memory totals, disk usage, and the full database filesystem path. When `MCP_ALLOW_ANONYMOUS_ACCESS=true` is set (required for the HTTP server to function without OAuth/API key), this endpoint is accessible without authentication. Combined with the default `0.0.0.0` binding, this exposes sensitive reconnaissance data to the entire network.\n\n### Details\n### Vulnerable Code\n\n**`health.py:90-101` - System information collection**\n\n```python\nsystem_info = {\n \"platform\": platform.system(), # e.g., \"Linux\", \"Darwin\"\n \"platform_version\": platform.version(), # Full OS kernel version string\n \"python_version\": platform.python_version(),# e.g., \"3.12.1\"\n \"cpu_count\": psutil.cpu_count(), # CPU core count\n \"memory_total_gb\": round(memory_info.total / (1024**3), 2),\n \"memory_available_gb\": round(memory_info.available / (1024**3), 2),\n \"memory_percent\": memory_info.percent,\n \"disk_total_gb\": round(disk_info.total / (1024**3), 2),\n \"disk_free_gb\": round(disk_info.free / (1024**3), 2),\n \"disk_percent\": round((disk_info.used / disk_info.total) * 100, 2)\n}\n```\n\n**`health.py:131-132` - Database path disclosure**\n\n```python\nif hasattr(storage, \u0027db_path\u0027):\n storage_info[\"database_path\"] = storage.db_path # Full filesystem path\n```\n\n### Authentication Bypass Path\n\nThe `/api/health/detailed` endpoint uses `require_read_access` which calls `get_current_user`. When `MCP_ALLOW_ANONYMOUS_ACCESS=true`, the auth middleware grants access:\n\n```python\n# middleware.py:372-379\nif ALLOW_ANONYMOUS_ACCESS:\n logger.debug(\"Anonymous access explicitly enabled, granting read-only access\")\n return AuthenticationResult(\n authenticated=True,\n client_id=\"anonymous\",\n scope=\"read\",\n auth_method=\"none\"\n )\n```\n\n**Note**: The basic `/health` endpoint (line 68) has **no auth dependency at all** and returns version and uptime information unconditionally.\n\n### Information Exposed\n\n| Field | Example Value | Reconnaissance Value |\n|-------|--------------|---------------------|\n| `platform` | `\"Linux\"` | OS fingerprinting |\n| `platform_version` | `\"#1 SMP PREEMPT_DYNAMIC...\"` | Kernel version \u2192 CVE targeting |\n| `python_version` | `\"3.12.1\"` | Python CVE targeting |\n| `cpu_count` | `8` | Resource enumeration |\n| `memory_total_gb` | `32.0` | Infrastructure profiling |\n| `database_path` | `\"/home/user/.mcp-memory/memories.db\"` | Username + file path disclosure |\n| `database_size_mb` | `45.2` | Data volume estimation |\n\n### Attack Scenario\n\n1. Attacker scans the local network for services on port 8000\n2. Finds mcp-memory-service with HTTP enabled and anonymous access\n3. Calls `GET /api/health/detailed` (no credentials needed)\n4. Receives OS version, Python version, full database path (revealing username), system resources\n5. Uses this information to:\n - Target known CVEs for the specific OS/Python version\n - Identify the database file location for potential direct access\n - Profile the system for further attacks\n \n### PoC\n\n```python\n# Show the system info that would be exposed\nimport platform, psutil\n\nsystem_info = {\n \"platform\": platform.system(),\n \"platform_version\": platform.version(),\n \"python_version\": platform.python_version(),\n \"cpu_count\": psutil.cpu_count(),\n \"memory_total_gb\": round(psutil.virtual_memory().total / (1024**3), 2),\n}\nprint(system_info) # All of this is returned to unauthenticated users\n```\n\n### Impact\n- **OS fingerprinting**: Exact OS and kernel version enables targeted exploit selection\n- **Path disclosure**: Database path reveals username, home directory structure, and file locations\n- **Resource enumeration**: CPU, memory, and disk info reveal infrastructure scale\n- **Reconnaissance enablement**: Combined information significantly reduces attacker effort for follow-up attacks\n\n## Remediation\n\n1. **Remove system details from default health endpoint** - return only `status`, `version`, `uptime`:\n\n```python\n@router.get(\"/health/detailed\")\nasync def detailed_health_check(\n storage: MemoryStorage = Depends(get_storage),\n user: AuthenticationResult = Depends(require_write_access) # Require admin/write access\n):\n # Only return storage stats, not system info\n ...\n```\n\n2. **Do not expose `database_path`** - this leaks the filesystem structure:\n\n```python\n# Remove or redact\n# storage_info[\"database_path\"] = storage.db_path # REMOVE THIS\n```\n\n3. **Add auth to basic `/health`** or limit it to status-only (no version):\n\n```python\n@router.get(\"/health\")\nasync def health_check():\n return {\"status\": \"healthy\"} # No version, no uptime\n```\nAlternatively, **Bind to `127.0.0.1` by default** instead of `0.0.0.0`, preventing network-based reconnaissance entirely:\n\n```python\n# In config.py \u2014 change default from \u00270.0.0.0\u0027 to \u0027127.0.0.1\u0027\nHTTP_HOST = os.getenv(\u0027MCP_HTTP_HOST\u0027, \u0027127.0.0.1\u0027)\n```\n\nUsers who need network access can explicitly set `MCP_HTTP_HOST=0.0.0.0`, making the exposure a conscious opt-in rather than a default.",
"id": "GHSA-73hc-m4hx-79pj",
"modified": "2026-03-09T15:50:08Z",
"published": "2026-03-05T21:42:10Z",
"references": [
{
"type": "WEB",
"url": "https://github.com/doobidoo/mcp-memory-service/security/advisories/GHSA-73hc-m4hx-79pj"
},
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2026-29787"
},
{
"type": "WEB",
"url": "https://github.com/doobidoo/mcp-memory-service/commit/18f4323ca92763196aa2922f691dfbeb6bd84e48"
},
{
"type": "PACKAGE",
"url": "https://github.com/doobidoo/mcp-memory-service"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:L/I:N/A:N",
"type": "CVSS_V3"
}
],
"summary": "mcp-memory-service Vulnerable to System Information Disclosure via Health Endpoint"
}
Drag nodes to arrange them. Double-click a related vulnerability to open it, or a source to visit its link. The Physics button on the left tunes the layout and the simulation; the legend at the bottom left hides or shows nodes by kind and by sighting type.
Sightings
| Author | Source | Type | Date | Other |
|---|
Nomenclature
- Seen: The vulnerability was mentioned, discussed, or observed by the user.
- Confirmed: The vulnerability has been validated from an analyst's perspective.
- Published Proof of Concept: A public proof of concept is available for this vulnerability.
- Exploited: The vulnerability was observed as exploited by the user who reported the sighting.
- Patched: The vulnerability was observed as successfully patched by the user who reported the sighting.
- Not exploited: The vulnerability was not observed as exploited by the user who reported the sighting.
- Not confirmed: The user expressed doubt about the validity of the vulnerability.
- Not patched: The vulnerability was not observed as successfully patched by the user who reported the sighting.
Vulnerability timeline
Sightings and recorded dates, grouped by day · UTC
How durations are calculated
Following the VTEM framework (Black Hat USA 2015), time to patch starts when the vendor or a coordinator was informed, total time to patch at discovery, and disclosure to patch at public disclosure; vendor response time runs from the report to its acknowledgement. Time to exploit and time unpatched start at the earliest discovery or report, falling back to public disclosure. Unpatched time ends at the first recorded patch, or the calculation time when no patch is recorded. Exploitation to patch shows the interval between those two events; a published proof of concept is a public exploit, not exploitation.
Each result shows its source dates. A negative interval means the end event came first; when date-only inputs leave the order open, the result says so. Date-only inputs are approximate. The earliest evidence wins: KEV report dates and sightings are labelled as such. Records describing another vulnerability are listed but never calculated from, and a PoC alone does not establish exploitation. Missing patch evidence does not prove that no patch exists. Filters below change the event list only.
Loading timeline…
The approach is described in our paper Mapping CVEs to MITRE ATT&CK Techniques: A Curated Gold-Set Classifier and the Limits of LLM-Assisted Label Expansion.
Browse all ATT&CK techniques and the vulnerabilities related to each.
Related by attack behaviour
Vulnerabilities whose description is nearest to this one in the vector space of the CIRCL/vulnerability-attack-technique-biencoder model. This is a similarity search over the bi-encoder space (plain cosine), not a classification, and it has no measured accuracy.