MCP Security Center. Use MCP without the surprises.
Plain-English guides to the risks of connecting AI apps to real systems through MCP, and the checklists, templates and controls that reduce them.
We grow your business with
- CROturn traffic into leads
- SEOrank & get found
- GEOwin generative search
- AEOget cited by AI answers
Every threat page explains the attack, who is at risk and how to reduce it, with sources: the MCP specification, OWASP and the researchers who found each issue. Example scenarios are labeled as illustrative unless they are real, sourced incidents.
Threats
Attacks on tools and trust
How attackers abuse tool descriptions, tool results, permissions and the servers you install.
Command Injection in Local MCP Servers
How command injection turns a local MCP server into a way to run code on your machine, three published CVEs, how to spot unsafe code and the fixes that work.
Read the guide →Data Exfiltration Through MCP
How data leaves an MCP setup: tool arguments, outbound tools, chat history and over-shared context. Real research cases, warning signs and controls that work.
Read the guide →Excessive Permissions in MCP Servers
Why over-broad scopes make every MCP attack worse, how permissions creep over time, how to spot it, and the least-privilege model the MCP spec recommends.
Read the guide →MCP Rug Pull Attacks: When a Trusted Server Changes
An MCP rug pull is when a server you approved changes its tools later. Learn how it works, how to spot it, and how pinning and change alerts reduce the risk.
Read the guide →MCP Supply Chain Risks: Malicious and Typosquatted Servers
How fake, typosquatted and backdoored MCP servers reach your machine, the postmark-mcp case, what the MCP Registry checks, and how to vet servers first.
Read the guide →MCP Tool Poisoning: What It Is and How to Prevent It
MCP tool poisoning hides instructions in tool descriptions that the AI reads but you don't see. Learn how it works, who is at risk, and how to reduce it.
Read the guide →MCP Tool Shadowing and Name Collisions
Tool shadowing lets one MCP server change how your AI uses another server's tools. Learn how it works, how name collisions add risk, and how to prevent both.
Read the guide →Prompt Injection Through MCP Tool Results
How hidden instructions in emails, issues and web pages reach your AI through MCP tool results, with a real GitHub case, warning signs and defenses that work.
Read the guide →The Confused Deputy Problem in MCP
How a confused deputy attack lets an attacker reuse a user's consent on an MCP proxy server, who is at risk, and the per-client consent the spec requires.
Read the guide →Why MCP Servers Must Not Pass Tokens Through
Token passthrough is forbidden by the MCP spec. Learn what it is, why it breaks audience checks and audit trails, how to spot it, and the right pattern to use.
Read the guide →
Threats
Network, data and operational risks
Local servers, secrets, sign-in mistakes, runaway costs and approval fatigue, plus the official guidance.
Approval Fatigue: When Users Click Yes to Everything
Why constant MCP approval prompts train people to click yes without reading, and how to design fewer, better prompts with allow and deny rules and sandboxing.
Read the guide →Common MCP OAuth Misconfigurations
The OAuth mistakes that most often break MCP security, from skipped audience checks to loose redirect URIs, mapped to the 2026-07-28 spec rules that stop them.
Read the guide →Cross-Server Data Leakage in MCP
How data from one MCP server ends up in another when an AI agent connects to several at once, with sourced research examples and practical ways to split them.
Read the guide →DNS Rebinding and Local MCP Servers
How a web page can reach an MCP server on your own computer through DNS rebinding, with real SDK advisories and the spec's fixes: Origin checks and auth.
Read the guide →Exposed Local MCP Servers
Why MCP servers bound to 0.0.0.0 or left without a login are reachable by others, what researchers found in the wild, and how to lock local servers down.
Read the guide →MCP Security Best Practices From the Spec
A plain-English tour of the official MCP Security Best Practices page for the 2026-07-28 spec: each attack it covers, the MUST rules, and who must act on them.
Read the guide →Resource Exhaustion and Runaway Costs in MCP
How MCP servers get overwhelmed and AI agents run up bills through loops and floods, with real advisories and the spec's rules on rate limits and timeouts.
Read the guide →Secrets and Credential Leakage in MCP
How API keys and OAuth tokens leak through MCP config files, logs, tool results and local helpers, with sourced examples and the spec's rules to prevent it.
Read the guide →Server-Side Request Forgery (SSRF) in MCP
How server-side request forgery hits MCP clients and servers, a labeled example, real advisories, and the spec's fixes: HTTPS, IP blocking, egress proxies.
Read the guide →The OWASP MCP Top 10, Explained
All ten OWASP MCP Top 10 risks in plain English, as listed on the beta project page, with what each means for your business and what the MCP spec says.
Read the guide →
Put it into practice
Checklists, templates and response
Copyable checklists, questionnaires and policies, a sourced incident tracker, and what to do if something goes wrong.
AI and MCP Acceptable Use Policy Template
A copyable AI and MCP acceptable use policy template: approved tools, data rules, MCP connections, approvals, credentials, logging and incident reporting.
Read the guide →How to Security-Review an MCP Server Before You Install It
A 9-step security review for any MCP server before you install it: publisher, tools, install command, transport, auth, scopes, advisories and a trial run.
Read the guide →MCP Audit Logging: What to Record
What to record in MCP audit logs: the fields for every tool call, scope and approval events, what to mask, how to protect logs, and what the spec and OWASP say.
Read the guide →MCP Incident Response: What to Do If a Server Is Compromised
What to do if an MCP server is compromised: warning signs, the first hour, revoking tokens, preserving logs, scoping the damage and scenario playbooks.
Read the guide →MCP Security Checklist (Copy and Use)
A copyable MCP security checklist: inventory, sign-in and scopes, local servers, approvals, logging, supply chain and reviews, tied to the spec and OWASP.
Read the guide →MCP Security Incidents and Vulnerabilities: A Sourced Tracker
A sourced tracker of MCP security incidents and CVEs from 2025 to October 2026: what happened, who was affected, the fix, and a link to the primary source.
Read the guide →MCP Security for Small Businesses
MCP security for small businesses in plain English: the risks that matter, ten practical rules, a one-page staff policy and what to do if something goes wrong.
Read the guide →MCP Vendor Security Questionnaire
A copyable MCP vendor security questionnaire: 35 questions on tokens, scopes, transports, tool changes, logging and patching, and how to score answers.
Read the guide →Sandboxing Local MCP Servers
How to sandbox local MCP servers: what the spec recommends, container and OS sandbox options, limits on files and network, and the gaps sandboxes do not cover.
Read the guide →
FAQ
Common questions
Is MCP safe to use?
What is the biggest MCP security risk?
Is there an official MCP security standard?
Can eSEOspace review our setup?
Project Managers who will work with you on your project!
Irina S.
Benjamin G.
David G.
Jeanette M.
Mark S.
Maryna P.
Want a second opinion on your MCP setup?
Request a free MCP readiness review. We'll look at the servers and AI apps you use and tell you what to lock down first.
Request a Free Review →
