| AAK-AGENT-ZERO-CVE-2026-30624-PIN-001 |
HIGH |
Agent Zero 0.9.8 MCP-config RCE (CVE-2026-30624, git-only pin) |
| AAK-AGENTSCOPE-CVE-2026-123023-001 |
HIGH |
AgentScope 2.0.0 to 2.0.8: the Bash tool counts tee as read-only, so an agent writes files without asking |
| AAK-ANTHROPIC-SDK-001 |
HIGH |
MCP server built on the upstream SDK without STDIO sanitizer |
| AAK-ASTROMCP-SQLI-CVE-2026-7591-001 |
HIGH |
astro-mcp-server SQL injection (CVE-2026-7591, npm 1.1.1 or earlier) |
| AAK-CHATCHAT-CVE-2026-75128-001 |
CRITICAL |
Langchain-Chatchat, every release and image (no fix): POST /tools/call runs the shell tool for an unauthenticated caller |
| AAK-CHATGPT-MCP-CVE-2026-7061-PIN-001 |
HIGH |
chatgpt-mcp-server OS command injection (CVE-2026-7061, npm/git 0.1.0 or earlier) |
| AAK-CLAUDECODE-CVE-2026-40068-PIN-001 |
HIGH |
Anthropic Claude Code below 2.1.251: folder-trust bypass (CVE-2026-40068), write-time symlink following (CVE-2026-103435, CVE-2026-88968) and plugin commit-pin bypass (CVE-2026-86063) |
| AAK-DNS-REBIND-002 |
HIGH |
Vulnerable MCP SDK version pinned (DNS-rebinding fix missing) |
| AAK-DOCSGPT-MCP-STDIO-MITM-001 |
HIGH |
DocsGPT MCP transport-flip MITM (OX 2026-05-01, CVE-2026-26015 family) |
| AAK-DORIS-001 |
HIGH |
apache-doris-mcp-server below 0.6.1 SQL injection |
| AAK-EXCEL-MCP-001 |
CRITICAL |
excel-mcp-server 0.1.7 or earlier path traversal |
| AAK-FLOWISE-001 |
CRITICAL |
Flowise below 3.1.2 MCP adapter authenticated RCE |
| AAK-FLOWISE-CVE-2026-119978-001 |
HIGH |
Flowise from 1.4.6, every release (no fix, the project has ended): the custom function sandbox lets a caller write a file into an allowed package and run it |
| AAK-GHA-IMMUTABLE-001 |
MEDIUM |
Third-party GitHub Action not pinned by full commit SHA |
| AAK-GHA-PUBLISH-GATE-001 |
HIGH |
Workflow job publishes a package with no environment approval gate |
| AAK-GPTRESEARCHER-MCP-STDIO-MITM-001 |
HIGH |
GPT-Researcher MCP transport-flip MITM (OX 2026-05-01, CVE-2025-65720) |
| AAK-HOOK-RCE-003 |
HIGH |
Claude Code below 1.0.111 runs project code before the trust dialog (CVE-2025-59536) |
| AAK-KNOWN-VULN-001 |
CRITICAL |
A locked, pinned or installed release is affected by a published advisory |
| AAK-KNOWN-VULN-002 |
CRITICAL |
A launched MCP server, binary or image is a release a published advisory lists as affected |
| AAK-KNOWN-VULN-003 |
MEDIUM |
An MCP config launches, without pinning a release, a package a published advisory lists as affected |
| AAK-LANGCHAIN-001 |
HIGH |
Project depends on langchain-core below 1.2.22 (load_prompt path traversal) |
| AAK-LANGCHAIN-003 |
HIGH |
LangChain deserialization of untrusted data |
| AAK-LITELLM-CVE-2026-30623-PIN-001 |
HIGH |
LiteLLM pin floor for CVE-2026-30623 (below 1.83.7) |
| AAK-MALICIOUS-EXTENSION-001 |
CRITICAL |
An agent host's installed editor extension is a build listed as malware |
| AAK-MALICIOUS-RELEASE-001 |
CRITICAL |
A declared, locked, installed or deployed release is listed as malware |
| AAK-MALICIOUS-RELEASE-002 |
HIGH |
An MCP config launches a package with releases listed as malware, without pinning a safe one |
| AAK-MARKETPLACE-001 |
HIGH |
Unsigned marketplace.json manifest |
| AAK-MARKETPLACE-002 |
MEDIUM |
Plugin permission set grants broad access |
| AAK-MARKETPLACE-003 |
HIGH |
Plugin name typosquats a well-known package |
| AAK-MARKETPLACE-004 |
HIGH |
Plugin source pins to a mutable git ref |
| AAK-MARKETPLACE-005 |
LOW |
Marketplace manifest could not be read |
| AAK-MCP-1MCP-CVE-2026-108586-001 |
MEDIUM |
1MCP Agent from 0.20.0: a negated tag filter reaches servers outside an OAuth token's tag scopes (no fixed release) |
| AAK-MCP-9ROUTER-CVE-2026-46339-001 |
CRITICAL |
9Router unauthenticated MCP bridge → command execution (below 0.5.2) |
| AAK-MCP-AGENTICFLOW-CVE-2026-58195-001 |
HIGH |
Agentic-Flow MCP tool params → execSync command injection (below 2.0.14) |
| AAK-MCP-AGENTICMAIL-CVE-2026-57495-001 |
HIGH |
AgenticMail bridge-wake indirect prompt injection (unpinned @agenticmail/* below fix) |
| AAK-MCP-AGENTSCOPE-CVE-2026-123022-001 |
HIGH |
AgentScope 2.x: the service takes a caller-set X-User-ID header as identity, and its workspace API starts any stdio MCP command (no fixed release) |
| AAK-MCP-AISDLC-CVE-2026-117558-001 |
HIGH |
ai-sdlc MCP server below 0.32.0: the repo argument reaches a gh command string |
| AAK-MCP-AMAZONMQ-CVE-2026-18655-001 |
MEDIUM |
Amazon MQ MCP Server broker-hostname SSRF exfiltrates credentials/tokens (below 2.0.24) |
| AAK-MCP-APACHEANSWER-CVE-2026-71563-001 |
MEDIUM |
Apache Answer image 2.0.0 to 2.0.2: the MCP answer tool returns answers to hidden, deleted and pending questions |
| AAK-MCP-APIFY-CVE-2026-46341-001 |
HIGH |
Apify MCP host-validation defects: docs-allowlist SSRF and Actor path-authority token leak (@apify/actors-mcp-server below 0.10.11) |
| AAK-MCP-APISIX-CVE-2026-124726-001 |
MEDIUM |
Apache APISIX image 3.19.0: an openapi-to-mcp tool call can reach API paths the OpenAPI document never declared, with the Route's credentials |
| AAK-MCP-APPIUM-CVE-2026-58500-001 |
HIGH |
MCP Appium locator-UI HTML/JS injection → unauthorized tool exec (below 1.85.10) |
| AAK-MCP-ARGOCD-CVE-2026-108585-001 |
MEDIUM |
argocd-mcp builds Argo CD API paths from tool arguments, so delete_application can reach repositories and clusters (no fixed release) |
| AAK-MCP-ASTRBOT-CVE-2026-15501-001 |
MEDIUM |
AstrBot MCP-test endpoint SSRF (affected up to 4.25.2) |
| AAK-MCP-ATLASSIAN-CVE-2026-124503-001 |
HIGH |
mcp-atlassian 0.22.0 to 0.24.0: the attachment upload guard is the working directory, so a wide one passes any file |
| AAK-MCP-ATLASSIAN-CVE-2026-27825-001 |
CRITICAL |
mcp-atlassian below 0.17.0 (CVE-2026-27825), or Jira/Confluence content reaching an exec sink |
| AAK-MCP-ATLASSIAN-CVE-2026-27826-001 |
HIGH |
mcp-atlassian below 0.17.0 (CVE-2026-27826), or Jira/Confluence content reaching a file-write sink |
| AAK-MCP-ATLASSIAN-CVE-2026-73498-001 |
HIGH |
mcp-atlassian below 0.22.0 (confluence_upload_attachment arbitrary file read) |
| AAK-MCP-ATOMICAGENTS-CVE-2026-91988-001 |
HIGH |
atomic-agents-stack below 1.1.0 (cleartext registry catalog reaches a local subprocess) |
| AAK-MCP-AUTHFETCH-CVE-2026-49857-001 |
HIGH |
auth-fetch-mcp 3.0.1 or earlier (IPv4-mapped IPv6 loopback bypasses the SSRF guard) |
| AAK-MCP-AWSAPIMCP-CVE-2026-115413-001 |
HIGH |
AWS API MCP Server 1.3.0 to 1.5.x: OAuth mode accepts tokens issued to other applications |
| AAK-MCP-AWSAPIMCP-CVE-2026-16584-001 |
HIGH |
AWS API MCP Server security-policy bypass on init failure (from 0.2.13 below 1.3.47) |
| AAK-MCP-AWSPOSTGRES-CVE-2026-85787-001 |
CRITICAL |
awslabs postgres-mcp-server: COPY TO PROGRAM command injection and SQL validation gap (below 1.1.7) |
| AAK-MCP-AWSSECAGENT-CVE-2026-87913-001 |
MEDIUM |
awslabs security-agent-mcp-server: scan output written to an unverified S3 bucket (below 0.2.0) |
| AAK-MCP-BETTERAUTH-CVE-2026-53512-001 |
HIGH |
Better Auth OAuth/MCP plugin flaw cluster — token bypass, code replay, weak crypto, open redirect (below 1.6.13) |
| AAK-MCP-BROWSEMCP-CVE-2026-55557-001 |
HIGH |
browse-mcp writes fetched content to a caller-controlled path (below 0.8.2) |
| AAK-MCP-CCTEMPLATES-CVE-2026-73222-001 |
HIGH |
claude-code-templates below 1.29.4 (--studio server: unauth 0.0.0.0:3444 → command-injection RCE) |
| AAK-MCP-CHROMEBRIDGE-CVE-2026-102878-001 |
HIGH |
mcp-chrome-bridge, every release (no fix): the native server's HTTP MCP endpoint answers web pages |
| AAK-MCP-CKAN-CVE-2026-73846-001 |
MEDIUM |
@aborruso/ckan-mcp-server below 0.4.112 (cache-key collision, prefix-only host check, verbose errors) |
| AAK-MCP-CLAUDEAGENTSDK-CVE-2026-96620-001 |
CRITICAL |
Claude Agent SDK below 0.2.121 (TypeScript below 0.3.208): a session id passed to resume can inject Claude CLI flags that run a command |
| AAK-MCP-CLAUDECODE-CVE-2026-103012-001 |
LOW |
Claude Code 2.0.68 to 2.1.259 fetched managed settings with a stored API key ahead of the Enterprise or Team sign-in |
| AAK-MCP-CLINE-CVE-2026-59723-001 |
HIGH |
Cline below 3.0.30 (Hub dashboard WebSocket origin bypass → RCE) |
| AAK-MCP-CLOUDRUN-CVE-2026-85297-001 |
HIGH |
Cloud Run MCP below 1.11.0: source deploys use a bucket named after the project without checking who owns it |
| AAK-MCP-CODEEXEC-SANDBOX-001 |
CRITICAL |
FastMCP's GenerativeUI runs caller code in a prefab-ui sandbox with open Deno permissions |
| AAK-MCP-CODEGRAPHRAG-CVE-2026-77604-001 |
HIGH |
Code Graph RAG below 0.0.345: the MCP read_file tool's paginated branch reads outside the project root |
| AAK-MCP-CODEGRAPHRAG-CVE-2026-87008-001 |
HIGH |
Code Graph RAG 0.0.421 to 0.0.584: MCP structural tools follow symlinks out of the project root |
| AAK-MCP-CODEWHALE-CVE-2026-75858-001 |
HIGH |
CodeWhale auto-approved code execution: rlm_eval and exec_shell_interact bypass the approval policy (codewhale from 0.8.41 below 0.8.64) |
| AAK-MCP-CONDUCTOR-CVE-2026-109402-001 |
MEDIUM |
Conductor image from 3.22.0: the LIST_MCP_TOOLS task connects to any mcpServer URL (SSRF, no fixed release) |
| AAK-MCP-CONTENTFUL-CVE-2026-53957-001 |
HIGH |
@contentful/mcp-server below 1.7.19 / @contentful/mcp-tools below 0.4.5 (LLM-controlled host and proxy carry the management token) |
| AAK-MCP-CONTEXTFORGE-CVE-2026-110279-001 |
HIGH |
IBM ContextForge MCP Gateway 0.8.0 to 1.1.0: X-Upstream-Authorization replaces the gateway's upstream credential, identity headers skip the per-gateway setting, jq guard DoS |
| AAK-MCP-CONTEXTFORGE-CVE-2026-18484-001 |
MEDIUM |
IBM ContextForge MCP Gateway 0.7.0 to 1.0.11: tool-name collision, cross-tenant import, DCR SSRF, token-exchange cache reuse, Admin UI XSS and schema ReDoS, fixed in 1.1.0 |
| AAK-MCP-CONTEXTFORGE-CVE-2026-77822-001 |
CRITICAL |
IBM ContextForge MCP Gateway: default credentials and JWT key, DNS-rebind and OAuth token-URL SSRF, jq-filter, Admin UI XSS and cross-session leaks (1.0.8 or earlier) |
| AAK-MCP-CORTEX-CVE-2026-49986-001 |
HIGH |
neuro-cortex-memory 3.17.0 or earlier (CLAUDE_PROJECT_DIR treated as a trusted source root) |
| AAK-MCP-D365FO-CVE-2026-114123-001 |
HIGH |
D365 F&O MCP Server below 1.11.0: the documented Azure deployment serves /mcp unauthenticated by default |
| AAK-MCP-DAGU-CVE-2026-102943-001 |
MEDIUM |
Dagu 2.7.0 to 2.10.7: the MCP dagu_read dag_spec target reads YAML files outside the DAG directories |
| AAK-MCP-DBTMCP-CVE-2026-44968-001 |
MEDIUM |
dbt-mcp flag injection + tool-arg leakage (dbt-mcp below 1.17.1) |
| AAK-MCP-DEEPSEEK-CVE-2026-55604-001 |
HIGH |
@arikusi/deepseek-mcp-server below 1.8.0 (unbound session IDs + unauth HTTP transport) |
| AAK-MCP-DIFY-CVE-2026-105761-001 |
HIGH |
Dify below 1.16.0: a workspace member can change another app's MCP server |
| AAK-MCP-DJUST-CVE-2026-113731-001 |
LOW |
djust 0.9.0 to 1.2.3: the find_handlers_for_template MCP tool reads any absolute path |
| AAK-MCP-DOCUMENTDB-CVE-2026-18954-001 |
MEDIUM |
AWS Labs DocumentDB MCP Server aggregation-pipeline authorization bypass (below 1.0.12) |
| AAK-MCP-DROIDMIND-CVE-2026-117559-001 |
HIGH |
DroidMind, every release: shell substitution passes the command allowlist |
| AAK-MCP-DYNATRACE-CVE-2026-123810-001 |
MEDIUM |
Dynatrace MCP server below 2.1.1: tool arguments reach DQL queries and workflow templates unescaped |
| AAK-MCP-DYNATRACE-CVE-2026-123811-001 |
HIGH |
Dynatrace MCP server 0.5.0 to 1.8.7: --http mode runs tool calls with no authentication |
| AAK-MCP-EDGEEVER-CVE-2026-109557-001 |
HIGH |
EdgeEver image below 1.113.0: notebook names, memo titles and templates written through MCP run as script (stored XSS) |
| AAK-MCP-ELVIX-CVE-2026-108542-001 |
MEDIUM |
@elvix.is/sdk MCP server below 0.10.2: a tool's path argument can send the elvix API key to any host |
| AAK-MCP-EZBOOKKEEPING-CVE-2026-86041-001 |
MEDIUM |
ezBookkeeping image 0.10.0 to 1.5.1: a spoofed X-Forwarded-For passes the MCP source-IP allowlist |
| AAK-MCP-FASTAGENT-CVE-2026-102610-001 |
MEDIUM |
fast-agent 0.3.18 to 0.7.20: fast-agent serve exposes the agent over MCP on every interface with no authentication |
| AAK-MCP-FASTMCPTELEGRAM-CVE-2026-55096-001 |
HIGH |
fast-mcp-telegram below 0.30.1 fetches attachment URLs its SSRF guard never resolved |
| AAK-MCP-FFMPEG-CVE-2026-117568-001 |
HIGH |
mcp-ffmpeg from 1.0.3: video paths reach a shell command string |
| AAK-MCP-FLORENCE2-CVE-2026-19984-001 |
MEDIUM |
mcp-florence2 SSRF via the get_images src argument (0.3.13 or earlier, no code fix planned) |
| AAK-MCP-FLOWISE-CVE-2026-91931-001 |
HIGH |
flowise below 3.1.4 (Custom MCP node reaches RCE through npx package names and an unvalidated cwd) |
| AAK-MCP-FLYTO-CVE-2026-67425-001 |
HIGH |
Flyto2 Core forwards provider API keys to a caller-controlled base_url, and its HTTP MCP route runs modules unauthenticated (below 2.26.6) |
| AAK-MCP-FROMOPENAPI-CVE-2026-115750-001 |
LOW |
mcp-from-openapi from 2.2.0 below 2.9.0: spec credentials follow a cross-origin redirect, and a trailing dot slips past blockedHosts |
| AAK-MCP-FROMOPENAPI-CVE-2026-59973-001 |
HIGH |
mcp-from-openapi below 2.5.0 (external $ref guard does not resolve or pin addresses — SSRF) |
| AAK-MCP-FRONTMCP-CVE-2026-67531-001 |
HIGH |
FrontMCP sandbox escape via Zod schema proxy reaches RCE (below 1.5.7) |
| AAK-MCP-FUNCTYPE-CVE-2026-59176-001 |
HIGH |
functype-mcp-server below 1.4.4: set_functype_version installs and imports any package |
| AAK-MCP-GATEWAY-REGISTRY-CVE-2026-100307-001 |
HIGH |
Amazon mcp-gateway-registry image 1.0.9 to 1.24.7: any signed-in user can map their group to the registry admin scope |
| AAK-MCP-GATEWAY-REGISTRY-CVE-2026-125926-001 |
HIGH |
Amazon mcp-gateway-registry image 1.0.3 to 1.34.0: a registered server's address can point the gateway at internal services |
| AAK-MCP-GATEWAY-REGISTRY-CVE-2026-14471-001 |
HIGH |
Amazon mcp-gateway-registry below 1.0.13 (SQL injection) |
| AAK-MCP-GEMINIBRIDGE-CVE-2026-54785-001 |
MEDIUM |
gemini-bridge tool-argument path traversal reads arbitrary files (from 1.0.0 below 1.3.1) |
| AAK-MCP-GENIEACS-CVE-2026-55637-001 |
HIGH |
genieacs-mcp serves an unauthenticated /mcp listener open to DNS rebinding (below 0.3.2) |
| AAK-MCP-GITLAB-ZEREIGHT-CVE-2026-61560-001 |
CRITICAL |
@zereight/mcp-gitlab below 2.1.30 (unauthenticated SSE + arbitrary file read, header-controlled API base, DNS rebinding) |
| AAK-MCP-GOOGLESEARCH-CVE-2026-19337-001 |
MEDIUM |
adenot mcp-google-search SSRF via the read_webpage url argument (0.3.1 or earlier, no fix released) |
| AAK-MCP-GRAFANA-CVE-2026-19516-001 |
CRITICAL |
Grafana mcp-grafana SSRF via caller-controlled X-Grafana-URL destination (below 1.1.0) |
| AAK-MCP-HEALTHLAKE-CVE-2026-15643-001 |
HIGH |
AWS HealthLake MCP server pagination SSRF → credential exfil (below 0.0.14) |
| AAK-MCP-HEALTHOMICS-CVE-2026-15415-001 |
MEDIUM |
AWS HealthOmics MCP workflow-bundle path traversal (below 0.0.36) |
| AAK-MCP-HELIO-CVE-2026-65957-001 |
MEDIUM |
Helio below 0.8.0: the dashboard sideband's CORS check admits hostnames that begin like a private address |
| AAK-MCP-HEYM-CVE-2026-100858-001 |
MEDIUM |
heym image below 0.0.109 fetches workflow-node URLs past its SSRF guard |
| AAK-MCP-HEYM-CVE-2026-67546-001 |
HIGH |
heym image below 0.0.75: the MCP fetch-tools endpoint starts any account's stdio command on the backend host |
| AAK-MCP-INSPECTOR-CVE-2026-23744-001 |
CRITICAL |
Vendored mcpjam-inspector fork carries CVE-2026-23744 |
| AAK-MCP-JAVASDK-CVE-2026-108395-001 |
HIGH |
MCP Java SDK 0.8.0 to 0.18.3, 1.0.0 to 1.1.3 and 2.0.0: the Servlet HTTP server transports read request bodies of any size |
| AAK-MCP-JSHOOK-CVE-2026-49856-001 |
MEDIUM |
@jshookmcp/jshook 0.3.1 (ICMP/traceroute tools bypass the SSRF authorization policy) |
| AAK-MCP-JUNIE-CVE-2026-111091-001 |
HIGH |
JetBrains Junie CLI before 26.8.3 (npm build below 2548.5.0): an untrusted project's MCP server configuration runs code |
| AAK-MCP-K6-CVE-2026-89039-001 |
MEDIUM |
Grafana mcp-k6 0.3.0 to 0.6.1: the convert_playwright_script prompt reads any file |
| AAK-MCP-K8S-CVE-2026-61459-001 |
CRITICAL |
mcp-server-kubernetes below 3.9.0 (argument injection → kubectl --server redirect) |
| AAK-MCP-KIMICODE-CVE-2026-95660-001 |
MEDIUM |
Kimi Code up to 0.31.0 starts a workspace's MCP servers before the trust prompt |
| AAK-MCP-KNOWNS-CVE-2026-86439-001 |
HIGH |
knowns MCP doc/memory and code.find tools: path traversal outside the project directory (no fixed release) |
| AAK-MCP-KORTIX-CVE-2026-108115-001 |
MEDIUM |
Kortix API image 0.10.7 to 0.13.51: IPv6 6to4 and Teredo addresses get past the SSRF guard |
| AAK-MCP-KOTLINSDK-CVE-2026-108714-001 |
HIGH |
MCP Kotlin SDK from 0.7.3: Application.mcpWebSocket accepts WebSocket frames of any size (no fixed release) |
| AAK-MCP-LANGBOT-CVE-2026-54449-001 |
HIGH |
LangBot lets any authenticated user configure an STDIO MCP command (no fix) |
| AAK-MCP-LANGCHAIN-MONGO-CVE-2026-106119-001 |
MEDIUM |
LangChain.js MongoDB chat history through 1.3.0: an object session id reaches another user's conversation |
| AAK-MCP-LANGCHAIN-REDIS-CVE-2026-105799-001 |
LOW |
LangChain.js Redis vector store below 1.1.1: unescaped filter values widen a RediSearch query |
| AAK-MCP-LANGFLOW-CVE-2026-12940-001 |
CRITICAL |
Langflow MCP stdio launcher env-var-injection RCE (from 1.0.0 below 1.13.0) |
| AAK-MCP-LANGGRAPH-API-CVE-2026-118668-001 |
HIGH |
langgraph-api 0.1.2 to 0.14.x: a JS or TS auth handler's namespace rewrite is dropped |
| AAK-MCP-LANGGRAPH-API-CVE-2026-55235-001 |
MEDIUM |
langgraph-api below 0.10.0 (loopback webhook bypasses per-user authorization) |
| AAK-MCP-LANGGRAPH-CHECKPOINT-CVE-2026-71433-001 |
MEDIUM |
LangGraph Postgres and SQLite stores leak across tenant namespaces (postgres below 3.1.3, sqlite below 3.1.2) |
| AAK-MCP-LANGGRAPH-CLI-CVE-2026-115885-001 |
MEDIUM |
LangGraph CLI from 0.4.15 below 0.4.32: langgraph deploy --verbose prints the registry push token |
| AAK-MCP-LANGGRAPH-MONGO-CVE-2026-48121-001 |
MEDIUM |
LangGraph MongoDB checkpoint saver NoSQL injection leaks checkpoints across tenants (below 1.3.1) |
| AAK-MCP-LANGGRAPHSDK-CVE-2026-104873-001 |
HIGH |
LangGraph SDK 0.1.45 to 0.4.3: resource-scoped auth decorators ignore actions |
| AAK-MCP-LANGGRAPHSDK-CVE-2026-118667-001 |
MEDIUM |
LangGraph SDK 0.4.0 to 0.4.5: v3 stream helpers put raw ids in authenticated request paths |
| AAK-MCP-LETTA-CVE-2025-51482-001 |
HIGH |
Letta (MemGPT) 0.16.4 or earlier (four CVEs: /v1/tools/run RCE, file-URL handler, AST parser) |
| AAK-MCP-LINEAGE-STAINLESS-001 |
INFO |
Stainless-generator provenance / lineage (informational) |
| AAK-MCP-LITELLM-CVE-2026-114612-001 |
CRITICAL |
LiteLLM from 1.87.0 below its fixed releases: session tokens share the at-rest encryption context, so an internal user can become proxy admin |
| AAK-MCP-LITELLM-CVE-2026-59822-001 |
HIGH |
LiteLLM below 1.84.0 (MCP auth bypass + skills-archive path traversal) |
| AAK-MCP-MARIMO-CVE-2026-75149-001 |
HIGH |
marimo launches a notebook-supplied MCP server command on open (below 0.23.15) |
| AAK-MCP-MARK3LABS-FS-CVE-2026-79534-001 |
MEDIUM |
mark3labs mcp-filesystem-server image (every tag): writes follow a dangling symlink out of the allowed directories |
| AAK-MCP-MARKETPLACE-CONFIG-FETCH-001 |
CRITICAL |
MCP server config fetched from a marketplace URL and spawned |
| AAK-MCP-MCPHUB-CVE-2026-79748-001 |
CRITICAL |
@samanhappy/mcphub below 1.1.8 (MCP gateway authorization, SSRF and credential-leak defects) |
| AAK-MCP-MEMSERVICE-CVE-2026-50027-001 |
CRITICAL |
mcp-memory-service below 10.67.1 (/api/documents/* served with no authentication) |
| AAK-MCP-METAMCP-CVE-2026-79538-001 |
CRITICAL |
MetaMCP image up to 2.4.22: inspector proxy code execution and cross-tenant session dispatch (no fixed release) |
| AAK-MCP-MYSQLMCP-CVE-2026-59971-001 |
CRITICAL |
mysql-mcp-server below 0.4.2 (unauthenticated SSE transport reaches execute_sql) |
| AAK-MCP-N8N-CVE-2026-126386-001 |
MEDIUM |
n8n 1.120.0 to 1.123.88, 2.x below 2.45.5 and 2.46.0 (MCP OAuth open redirect, Browser Use upload tool reads the server's files) |
| AAK-MCP-N8N-CVE-2026-59207-001 |
MEDIUM |
n8n below 2.41.4 / 2.42.0 (webhook, form and MCP routes run any project's workflow without its identifier; MCP tool bypasses the credential domain allow-list) |
| AAK-MCP-N8N-CVE-2026-65594-001 |
HIGH |
n8n MCP Server Trigger OAuth workflow-authorization bypass (from 2.27.0 below 2.29.8, from 2.30.0 below 2.30.1) |
| AAK-MCP-N8N-CVE-2026-72768-001 |
HIGH |
n8n below 2.41.4 / 2.42.0 (MCP workflow-validation prototype pollution, MCP Client SSRF bypass, node-schema loader RCE, cross-project credential reference, inline sub-workflow credential exfiltration) |
| AAK-MCP-N8NMCP-CVE-2026-54052-001 |
CRITICAL |
n8n-MCP multi-tenant backup isolation bypass (n8n-mcp below 2.57.4) |
| AAK-MCP-NEOMJS-CVE-2026-18482-001 |
HIGH |
Neo.mjs file-system MCP server shell injection (no released fix yet) |
| AAK-MCP-NETBOX-CVE-2026-57207-001 |
MEDIUM |
NetBox MCP Server image below 1.2.1: the HTTP transport serves every NetBox tool without authentication |
| AAK-MCP-NETLICENSING-CVE-2026-54446-001 |
CRITICAL |
netlicensing-mcp below 0.1.8 (a path-traversed product number reads API tokens in plaintext; unauthenticated /mcp falls back to the operator's API key below 0.1.6) |
| AAK-MCP-NETWORKAI-CVE-2026-109331-001 |
HIGH |
network-ai 5.14.0 to 5.15.1: ClaudeHookBridge checks deny patterns against one truncated field |
| AAK-MCP-NETWORKAI-CVE-2026-109332-001 |
LOW |
network-ai 4.0.0 to 5.15.1: the MCP SSE server lets other origins read its unauthenticated endpoints |
| AAK-MCP-NEXTCLOUD-CVE-2026-55640-001 |
CRITICAL |
Nextcloud MCP Server webhook endpoint is unauthenticated by default (below 0.117.2) |
| AAK-MCP-NEXTJS-CVE-2026-94486-001 |
LOW |
Next.js 16.0.0 to 16.3.7: the next dev server's MCP endpoint answers cross-site requests (dev server only) |
| AAK-MCP-NEZHA-CVE-2026-111385-001 |
MEDIUM |
Nezha dashboard image v2.1.0 to v2.3.14: MCP file-transfer URLs built from the Host header |
| AAK-MCP-NGINXUI-CVE-2026-84317-001 |
HIGH |
Nginx UI image 2.0.1 to 2.4.0: MCP config tools write blocked nginx directives and skip the secure-session step-up (code execution where njs or Perl is loaded) |
| AAK-MCP-NULLHUB-CVE-2026-110616-001 |
HIGH |
NullHub image through v2026.5.29: the /api management routes accept requests with no authentication (no fixed release) |
| AAK-MCP-OBOT-CVE-2026-101084-001 |
CRITICAL |
Obot image below v0.25.0: MCP gateway access-control bypasses, open OAuth client registration, SSRF |
| AAK-MCP-OBOT-CVE-2026-105138-001 |
MEDIUM |
Obot image v0.12.0 to v0.26.1: a user granted a catalog entry can read its static secrets |
| AAK-MCP-OBSIDIANTC-CVE-2026-115477-001 |
HIGH |
obsidian-tc from 1.0.1 below 1.31.4: a declined confirmation can still run a destructive call, and memory tools overwrite notes they do not own |
| AAK-MCP-OBSIDIANTC-CVE-2026-77366-001 |
CRITICAL |
obsidian-tc below 1.3.1: any HTTP token reaches every configured vault, the HTTP transport answers cross-origin requests, and the read ACL leaks |
| AAK-MCP-ODOOMCP-CVE-2026-108861-001 |
MEDIUM |
odoo-mcp from 1.0.0: execute_method returns fields the field-level ACL denies (no fixed release) |
| AAK-MCP-OFFICEPPT-CVE-2025-71427-001 |
MEDIUM |
Office-PowerPoint-MCP-Server tools read and write any path they are given (no fixed release) |
| AAK-MCP-OMNIGENT-CVE-2026-62674-001 |
CRITICAL |
Omnigent shared-agent swap injects a stdio MCP command (below 0.3.0) |
| AAK-MCP-OPENCLAW-CVE-2026-102807-001 |
MEDIUM |
OpenClaw 2026.7.2 up to 2026.9.4: a read-scoped operator runs write-scoped MCP App tools |
| AAK-MCP-OPENCLAW-CVE-2026-62195-001 |
HIGH |
OpenClaw MCP authorization and network-policy bypasses (below 2026.7.1) |
| AAK-MCP-OPENCOMPUTERUSE-CVE-2026-108721-001 |
MEDIUM |
Open Computer Use on macOS: a case-variant bundle id passes the password-manager denylist (no fixed release) |
| AAK-MCP-OPENHUMAN-CVE-2026-54140-001 |
HIGH |
OpenHuman core image 0.54.0 to 0.57.11: remote MCP tool descriptions reach the model unfiltered |
| AAK-MCP-PAYLOAD-CVE-2026-105806-001 |
HIGH |
Payload MCP plugin 3.61.0 to 3.87.x: a user can manage another account's MCP API keys |
| AAK-MCP-PDFREADER-CVE-2026-62264-001 |
HIGH |
PDF Reader MCP 2.2.0 to 3.0.0: the HTTP transport never checks the configured API key |
| AAK-MCP-PENPOT-CVE-2026-100868-001 |
MEDIUM |
Penpot MCP plugin WebSocket bridge listens on every interface (below 2.18.0) |
| AAK-MCP-PENPOT-CVE-2026-45805-001 |
CRITICAL |
Penpot MCP ReplServer unauthenticated /execute RCE (below 2.15.0) |
| AAK-MCP-PGEDGE-CVE-2026-125320-001 |
HIGH |
pgEdge Postgres MCP Server image below 1.1.0: read-only bypass, SQL injection through pl-do tools, DNS rebinding, a spoofable rate-limit address and a count_rows oracle |
| AAK-MCP-PGEDGE-CVE-2026-126199-001 |
MEDIUM |
pgEdge Postgres MCP Server's CLI and web client images below 1.1.0: database writes run without the write-confirmation prompt |
| AAK-MCP-PGEDGE-PLPERL-001 |
HIGH |
pgEdge Postgres MCP Server image below 1.2.0: Perl code injection through the arguments of plperl and plperlu pl-do tools |
| AAK-MCP-PILLMWIKI-CVE-2026-102911-001 |
CRITICAL |
pi-llm-wiki below 0.11.8 runs a shell on the URL its MCP capture tool is given |
| AAK-MCP-POSTGRESMCP-CVE-2026-85620-001 |
HIGH |
Postgres MCP Pro restricted-mode bypass via FROM-clause functions (no fixed release) |
| AAK-MCP-PRAISONAI-CVE-2026-61427-001 |
HIGH |
PraisonAI MCP unauthenticated-default + path traversal (below 4.6.78) |
| AAK-MCP-PRAISONAI-TS-CVE-2026-57139-001 |
CRITICAL |
praisonai (npm, TypeScript) below 1.7.2 (MCP HTTP server binds unrestricted with no auth) |
| AAK-MCP-QWED-CVE-2026-55546-001 |
CRITICAL |
QWED-MCP evaluates a tool argument through SymPy parse_expr (below 0.2.1) |
| AAK-MCP-REMOTE-CVE-2026-108716-001 |
MEDIUM |
mcp-remote from 0.8.0 launched with --device-code: client secret and tokens over plain HTTP (no fixed release) |
| AAK-MCP-REMOTE-CVE-2026-51994-001 |
CRITICAL |
mcp-remote 0.1.16–0.1.38 trusts URLs a remote MCP server chooses (SSRF, browser launch) |
| AAK-MCP-REPOMIX-CVE-2026-49988-001 |
MEDIUM |
Repomix MCP server bypasses secret-scan file-read boundary (below 1.14.1) |
| AAK-MCP-RIPGREP-CVE-2026-113378-001 |
MEDIUM |
mcp-ripgrep hands a tool's fileType argument to a shell unescaped (no fixed release) |
| AAK-MCP-RUFLO-CVE-2026-59726-001 |
CRITICAL |
ruflo below 3.16.3 (unauthenticated MCP bridge → tools/call RCE) |
| AAK-MCP-SDK-CVE-2026-52869-001 |
HIGH |
MCP Python SDK session/task cross-client access and unbounded request bodies (mcp below 1.29.1, and 2.0.x) |
| AAK-MCP-SDK-CVE-2026-59951-001 |
MEDIUM |
MCP Python SDK 1.8.0 to 1.29.x and 2.0.x/2.1.x: unbounded sessions, credentials sent across redirects, audience-blind bearer auth and fetched schema $refs |
| AAK-MCP-SEARXNG-CVE-2026-58483-001 |
HIGH |
mcp-searxng below 1.7.1 (a missing Content-Length bypasses the read cap) |
| AAK-MCP-SEQTHINKING-CVE-2026-81845-001 |
MEDIUM |
mcp-sequential-thinking 0.5.0 or earlier (session import/export path traversal) |
| AAK-MCP-SFMCP-CVE-2026-117561-001 |
HIGH |
sf-mcp, every release: Salesforce CLI flag values reach a shell command string |
| AAK-MCP-SHARIQ-GITHUB-CVE-2026-102906-001 |
MEDIUM |
@0xshariq/github-mcp-server runs git commands built from tool arguments through a shell (no fixed release) |
| AAK-MCP-SIYUAN-CVE-2026-116168-001 |
MEDIUM |
SiYuan image from 3.7.0 below 3.8.7: the MCP asset tool copies .env files into the workspace, where /assets/ serves them back |
| AAK-MCP-SIYUAN-CVE-2026-123492-001 |
MEDIUM |
SiYuan image 3.8.0 to 3.8.8: an MCP resources/read of a missing resource crashes the kernel |
| AAK-MCP-SIYUAN-CVE-2026-126393-001 |
MEDIUM |
SiYuan image from 3.7.0 below 3.8.10: the MCP skill tool's install download follows a redirect or a rebinding name to internal addresses |
| AAK-MCP-STATA-CVE-2026-47708-001 |
HIGH |
stata-mcp below 1.19.0 (Stata command injection, two CVEs) |
| AAK-MCP-STDIO-CMD-INJ-001 |
CRITICAL |
MCP STDIO server launched from network-controlled input (Python) |
| AAK-MCP-STDIO-CMD-INJ-002 |
CRITICAL |
MCP StdioClientTransport built from network-controlled input (TypeScript) |
| AAK-MCP-STDIO-CMD-INJ-003 |
CRITICAL |
MCP StdioServerParameters built from network-controlled input (Java) |
| AAK-MCP-STDIO-CMD-INJ-004 |
CRITICAL |
MCP STDIO command spawned from network-controlled input (Rust) |
| AAK-MCP-STDIO-CMD-INJ-005 |
CRITICAL |
MCP STDIO command spawned from network-controlled input (Go) |
| AAK-MCP-STDIO-UNBOUNDED-BUFFER-001 |
MEDIUM |
MCP Kotlin SDK stdio read buffer grows without bound (CVE-2026-53937, 0.7.0–0.12.0) |
| AAK-MCP-SUBLINEAR-CVE-2026-55609-001 |
HIGH |
consciousness-explorer state tools accept a caller-controlled path (sublinear-time-solver below 1.6.0, consciousness-explorer below 1.1.2) |
| AAK-MCP-TEXTEDITOR-CVE-2026-15138-001 |
MEDIUM |
tumf mcp-text-editor path traversal (affected up to 1.0.2) |
| AAK-MCP-TOOLBOX-CVE-2026-102242-001 |
HIGH |
Google MCP Toolbox 1.2.0 to 1.9.0: local-path checks follow symlinks out of the allowed roots |
| AAK-MCP-TOOLBOX-CVE-2026-93328-001 |
CRITICAL |
Google MCP Toolbox 1.2.0 to 1.4.x: the Cloud Storage tools take any bucket and any local path |
| AAK-MCP-TOOLBOX-HOSTS-001 |
MEDIUM |
MCP Toolbox for Databases on a network listener with an open Host or Origin check |
| AAK-MCP-TOOLUNIVERSE-CVE-2026-81096-001 |
CRITICAL |
ToolUniverse unauthenticated RCE via python_code_executor sandbox escape (1.2.6 or earlier) |
| AAK-MCP-TRAWL-CVE-2026-122140-001 |
MEDIUM |
Trawl image 1.5.0 to 1.9.3: IPv4-mapped IPv6 addresses and DNS rebinding pass the MCP outbound URL policy |
| AAK-MCP-TRIGGERDEV-CVE-2026-114090-001 |
HIGH |
Trigger.dev CLI 4.0.0 to 4.5.3: trigger dev --mcp serves MCP on every interface with no authentication |
| AAK-MCP-TRINITY-CVE-2026-108756-001 |
MEDIUM |
Trinity backend image through 0.9.5: an agent's MCP key can replace or delete its owner's Telegram binding (no fixed release) |
| AAK-MCP-TSSDK-CVE-2026-104850-001 |
HIGH |
MCP TypeScript SDK 1.12.0 to 1.30.1 (2.x client below 2.2.0): the MCP server picks where OAuth credentials go |
| AAK-MCP-TSSDK-CVE-2026-111767-001 |
HIGH |
MCP TypeScript SDK 1.24.0 to 1.31.0: experimental tasks are not tied to the session that created them |
| AAK-MCP-TSSDK-CVE-2026-111768-001 |
HIGH |
MCP TypeScript SDK: one HTTP request can exhaust server memory (validation work, body size, batch length) |
| AAK-MCP-TSSDK-CVE-2026-111769-001 |
MEDIUM |
MCP TypeScript SDK client: HTTP transports and OAuth requests send headers and bodies across a cross-origin redirect |
| AAK-MCP-TSSDK-CVE-2026-111770-001 |
HIGH |
MCP TypeScript SDK server 2.0.0 to 2.2.0: a shared transport or server delivers one HTTP client's messages to another |
| AAK-MCP-TSSDK-CVE-2026-111771-001 |
MEDIUM |
MCP TypeScript SDK: server bearer authentication accepts a token issued for another service |
| AAK-MCP-UNICLI-CVE-2026-58497-001 |
HIGH |
Uni-CLI from 0.208.0 below 0.225.2: the legacy HTTP MCP transport answers a web page's request to loopback /mcp |
| AAK-MCP-UNLA-CVE-2026-108865-001 |
HIGH |
Unla gateway image through v0.10.0: the built-in OAuth2 server issues tokens without authenticating anyone (no fixed release) |
| AAK-MCP-UTCP-CVE-2026-101057-001 |
LOW |
utcp-mcp below 1.1.3 dials plain-HTTP MCP server URLs from a call template |
| AAK-MCP-VOICEMODE-CVE-2026-79535-001 |
MEDIUM |
VoiceMode below 8.10.2 writes MCP-supplied config values into a sourced env file unescaped |
| AAK-MCP-WHATSAPP-CVE-2026-46555-001 |
HIGH |
WhatsApp MCP (verygoodplugins/whatsapp-mcp) below 0.2.1 (unauthenticated loopback bridge + media_path traversal → file exfil) |
| AAK-MCP-ZSCALER-CVE-2026-59563-001 |
MEDIUM |
Zscaler MCP Server 0.7.0 and 0.7.1: delete confirmation tokens are not bound to the resource |
| AAK-MCPCALC-CVE-2026-44717-PIN-001 |
CRITICAL |
MCP Calculate Server eval() RCE (CVE-2026-44717, PyPI below 0.1.1) |
| AAK-METAADS-CVE-2026-48039-001 |
CRITICAL |
Meta Ads MCP forwards unauthenticated requests and leaks the access token (below 1.0.109) |
| AAK-NEO4J-001 |
MEDIUM |
mcp-neo4j-cypher below 0.6.0 APOC read-only bypass |
| AAK-NETWORKAI-CVE-2026-120311-001 |
HIGH |
network-ai 5.0.0 to 5.15.4: FileJobStore job ids and AgentVCR cassette names escape their directory |
| AAK-OAUTH-3P-001 |
MEDIUM |
Repo depends on a third-party agent-platform SDK |
| AAK-PRAISONAI-CVE-2026-61438-001 |
HIGH |
PraisonAI 4.6.78 and later: a job workflow's inline Python step gets out of the executor's AST sandbox (no fixed release) |
| AAK-RUNTIME-ADV-001 |
CRITICAL |
Declared Node.js, PHP or Python runtime below the fixed release on its line |
| AAK-RUNTIME-ADV-002 |
INFO |
Runtime named by a floating tag on a line an advisory fixed |
| AAK-RUNTIME-ADV-003 |
CRITICAL |
Lockfile or exact pin puts a library bundling an affected codec below its fix |
| AAK-SEC-MD-001 |
LOW |
MCP server repo missing SECURITY.md or security_contact |
| AAK-SK-INMEMORY-VECTORSTORE-FILTER-CVE-2026-26030-PIN-001 |
CRITICAL |
Microsoft Semantic Kernel InMemoryVectorStore filter RCE (CVE-2026-26030, PyPI below 1.39.4) |
| AAK-SKILL-UNTRUSTED-EXEC-PATH |
HIGH |
Untrusted-search-path executable override in skill/install flow |
| AAK-STARTUP-HOOK-DOCKERFILE-001 |
MEDIUM |
Dockerfile adds a Python startup hook, or sets PYTHONSTARTUP in an MCP server image |
| AAK-STARTUP-HOOK-NPM-INSTALL-001 |
LOW |
npm lockfile entry with an install script in an MCP project |
| AAK-STARTUP-HOOK-PTH-001 |
HIGH |
Startup code in a site-packages .pth or .start file |
| AAK-STARTUP-HOOK-SITECUSTOMIZE-001 |
HIGH |
sitecustomize or usercustomize module where Python imports it at every start |
| AAK-SUPPLY-001 |
MEDIUM |
MCP server package not pinned to exact version |
| AAK-SUPPLY-002 |
HIGH |
Known vulnerable package in lockfile |
| AAK-SUPPLY-003 |
MEDIUM |
Dependency uses install scripts |
| AAK-SUPPLY-004 |
MEDIUM |
No lockfile present |
| AAK-SUPPLY-005 |
LOW |
Dependency count exceeds threshold |
| AAK-SUPPLY-006 |
HIGH |
Dependency with known MCP-specific vulnerability |