Latest Updates & News
Series: Threats to MCP servers
Risky deployments span public HTTP endpoints, shared npm packages, and opaque Docker images. These posts define each threat, how it shows up in MCP, and what changes once traffic flows through Guardian—plus how teams validate assumptions using the free MCP Playground, free MCP session risk calculator, and free MCP traffic monitor.
- Prompt injection driving tool abuse
- Secret exfiltration via arguments or responses
- Destructive shell and filesystem patterns
- SSRF and internal network access via tools
- Shadow IT MCP servers
- Supply chain and compromised MCP servers
- Data retention and compliance gaps
- Dangerous MCP tool sequences: real patterns that bypass ‘allowed tool’ lists
- Argument-level attacks on MCP: when the tool name is allowed but the payload is not
Stream Your MCP Audit Logs to Your Own Grafana, Sentry, or Webhook
MCP Trail now has log drains: forward every governed MCP tool call to your own Loki/Grafana, Sentry/GlitchTip, or an HTTPS webhook — tenant-scoped, encrypted, SSRF-guarded, with a one-click test.
MCP Trail Team
Product
Register WebMCP Tools in React with useModelTool
Register WebMCP tools with a React hook: useModelTool adds a tool on mount and removes it on unmount, so an agent only ever sees the tools for what's currently on screen.
MCP Trail Team
Developer Relations
Client-Side Approvals for WebMCP Tool Calls
Add a client-side human-in-the-loop step to WebMCP so an agent has to ask before a write tool runs. Approve agent tool calls in the browser with a per-tool policy — auto, confirm, or deny.
MCP Trail Team
Developer Relations
Debug WebMCP Tools with a Drop-in Devtools Panel
Debug and inspect WebMCP tools with a zero-config devtools panel. It lists every registered tool, auto-builds a form from each inputSchema, invokes it, and shows the result.
MCP Trail Team
Developer Relations
Detect WebMCP Toolset Drift on the Client
A WebMCP site can silently change the tools it exposes — that's drift. Detect WebMCP toolset drift in the browser: hash the registered toolset, freeze a baseline, and get notified the moment it changes.
MCP Trail Team
Developer Relations
Expose Your WebMCP Tools as an MCP Server
WebMCP tools live in the browser, so only a browser agent can call them. Bridge WebMCP to MCP: implement initialize, tools/list, and tools/call over any transport so any MCP client can reach them.
MCP Trail Team
Developer Relations
How to Test WebMCP Tools (Without a Flagged Chrome)
Test WebMCP tools as plain unit tests. A mock navigator.modelContext plus a driver turns tool calls into Vitest assertions — no flagged Chrome, no live model.
MCP Trail Team
Developer Relations
See What a WebMCP Agent Changed on Your Page
Visualize WebMCP agent actions. Wrap each tool call as a transaction, diff the DOM, and draw an overlay ring on exactly what an agent changed — with a timeline and screen-reader announcements.
MCP Trail Team
Developer Relations
Type-Safe WebMCP Tools with TypeScript
Define type-safe WebMCP tools in TypeScript with defineTool: it infers your execute argument types straight from the JSON Schema, so args are typed with zero hand-written interfaces.
MCP Trail Team
Developer Relations
Use WebMCP on Any Browser with a Polyfill
Use WebMCP without the Chrome flag: a polyfill installs navigator.modelContext only when it's missing, so your tools register on any browser and you can build and demo today.
MCP Trail Team
Developer Relations
How to Add WebMCP to a Website: Expose Your Site's Actions to AI Agents
A step-by-step WebMCP tutorial: register your site's actions as tools with navigator.modelContext so AI agents can call them in the browser. Code, testing, and best practices.
MCP Trail Team
Developer Relations
How to Enable Human-in-the-Loop (HITL) Approvals for WebMCP Tools
WebMCP tools run as the logged-in user with no confirmation step. Add human-in-the-loop approvals per tool with the Webmcp Trail extension and MCP Trail — sensitive calls pause until a person approves.
MCP Trail Team
Security Team
Turn Any WebMCP Site into a Guarded MCP Server with Auth and Logs
WebMCP tools run in the browser with no auth or audit trail. Use the Webmcp Trail extension + MCP Trail to bridge them into a governed MCP server — authentication, logging, policy, and approvals in a few clicks.
MCP Trail Team
Product Team
What Is WebMCP? How Websites Expose Tools to AI Agents
WebMCP lets a website publish its own actions as tools an AI agent can call right in the browser. Here's what WebMCP is, how it works, and why it matters — in plain English.
MCP Trail Team
Product Team
How to Add OAuth to an MCP Server (Without Writing Auth Code)
MCP servers now expect OAuth 2.1 — discovery, dynamic client registration, PKCE, token refresh. Get all of it on your existing MCP server in one click with MCP Trail.
MCP Trail Team
Security Team
AI Agent Firewall: What It Is, Where It Sits, and Your Options in 2026
AI agents execute real actions — so they need a firewall that inspects tool calls, not packets. The three architectures (model-side, host-side, protocol gateway), honest trade-offs, and how to deploy one.
MCP Trail Team
Security Team
How to Block Destructive Shell Commands from AI Agents (MCP Shell Safety)
AI agents with shell access can erase data with one hallucinated command. Turn on dangerous command blocking for MCP shell tools — a one-toggle guardrail in MCP Trail.
MCP Trail Team
Security Team
MCP Audit Logging: The Complete Guide (What to Log, Retention, Compliance)
Native MCP logging is fragmented and ephemeral — useless for security review or compliance. What a real MCP audit log records, field by field, and how to get one without building it.
MCP Trail Team
Security Team
MCP Fail-Open vs Fail-Closed: Choosing Your Enforcement Posture
What should happen to an AI tool call when your policy backend is unreachable — block it or let it through? Strict, Resilient, and Observe-only postures in MCP Trail, explained.
MCP Trail Team
Security Team
How to Rate-Limit MCP Tool Calls (Stop Runaway Agent Loops)
One retry loop can burn your API quota and your credits overnight. Set a token-bucket rate limit on MCP tool calls — RPM and burst — in the MCP Trail dashboard.
MCP Trail Team
Technical Team
How to Set Request & Response Size Limits for MCP Servers
Oversized payloads and deeply nested JSON can crash an MCP service — and giant responses can blow out your model's context. Configure both caps in MCP Trail, step by step.
MCP Trail Team
Technical Team
MCP Session Risk Scoring: Stop Runaway AI Agents Before They Do Damage
Per-call rules miss the real threat: sessions that go bad gradually. Learn how MCP session risk scoring works and how to enable it — monitor first, then block or require approval.
MCP Trail Team
Security Team
How to Control Which MCP Tools Your AI Can Run (Allow, Audit, Approve, Deny)
Every MCP server ships more tools than your agents need. Set a per-tool policy — allow, audit-only, human approval, or deny — in the MCP Trail dashboard, step by step.
MCP Trail Team
Security Team
How to Enforce Tool Order in MCP Sessions (Tool Sequence Policies)
Some tools should never run before others: no deploy before tests, no delete before backup. Enforce tool prerequisites per MCP session with sequence policies in MCP Trail.
MCP Trail Team
Security Team
How to Secure an Existing Remote MCP Server (Even One With No Auth)
Already have an MCP server running in the cloud — token-protected or wide open? Put a security gateway in front of its HTTP URL in five minutes: auth, policies, DLP, and audit.
MCP Trail Team
Security Team
How to Approve AI Tool Calls from Slack (MCP Approval Alerts)
Human-in-the-loop only works if humans see the request. Get Slack alerts with approval links the moment a risky MCP tool call is waiting for review — setup in minutes.
MCP Trail Team
Technical Team
How to Validate MCP Tool Arguments with JSON Schema (Input Contracts)
Models send malformed and manipulated tool arguments constantly. Enforce each tool's declared JSON Schema at the gateway — input contract validation in MCP Trail, step by step.
MCP Trail Team
Security Team
How to Check MCP Server Logs: Debug, Audit, and Monitor AI Tool Calls
MCP server not working, or wondering what your AI actually did? Step-by-step guide to reading MCP server logs, audit trails, and live traffic monitoring with MCP Trail.
MCP Trail Team
Technical Team
How to Host an MCP Server in the Cloud (No Docker, No Localhost)
Stop running MCP servers on your laptop. Host any MCP server in the cloud in minutes — no Docker, no VPS, no config files — with a free MCP Trail account. Step-by-step guide.
MCP Trail Team
Technical Team
How to Combine Multiple MCP Tools into One Workflow (Composite Tools)
Chain MCP tools into a single workflow the AI calls as one tool: fixed steps, sequential or parallel, no improvisation. Step-by-step guide to composite tools with MCP Trail.
MCP Trail Team
Technical Team
How to Protect Sensitive Data When Using MCP Servers (DLP for AI Tools)
MCP servers move real data through AI assistants — secrets, customer records, internal docs. Step-by-step guide to DLP rules, redaction, approvals, and audit with MCP Trail.
MCP Trail Team
Security Team
How to Combine Multiple MCP Servers into One Endpoint (MCP Bundles)
Too many MCP servers in your client config? Combine GitHub, Notion, Slack and more into a single MCP bundle: one endpoint, one token, one policy layer. Step-by-step with MCP Trail.
MCP Trail Team
Technical Team
How to Set Up the Jira MCP Server: AI Project Management Without the Token Sprawl
Set up the Jira MCP server the clean way: create a free MCP Trail account, connect Jira from the marketplace, and let AI triage issues and manage sprints with a full audit trail.
MCP Trail Team
Technical Team
How to Set Up the Slack MCP Server: Let AI Read Channels, Post Only With Approval
Set up the Slack MCP server without pasting a bot token into a config file: create a free MCP Trail account, connect Slack, allowlist read-only tools, and put every post behind human approval.
MCP Trail Team
Technical Team
How to Connect Claude Desktop & Cursor to a Hosted MCP Server
Connect Claude Desktop, Claude Code, Cursor, or VS Code to a remote MCP server in one config entry: endpoint URL + bearer token. Step-by-step guide with MCP Trail.
MCP Trail Team
Technical Team
How to Set Up the Notion MCP Server: Connect AI to Your Workspace Safely
Set up the Notion MCP server the safe way: create a free MCP Trail account, connect Notion from the marketplace, and give Claude or Cursor gated access to your pages and databases.
MCP Trail Team
Technical Team
How to Set Up the GitHub MCP Server Safely: Hosted, Governed, No Local Install
Set up the GitHub MCP server the safe way: connect it hosted on MCP Trail, put write actions like open/merge PR behind human approval, allowlist read-only tools, and audit every call.
MCP Trail Team
Technical Team
How to Use the Free MCP Playground, Session Risk Calculator, and Traffic Monitor
Step-by-step guide to run safer MCP tests with MCP Trail's free MCP Playground, free MCP session risk calculator, and free MCP traffic monitor.
MCP Trail Team
Security Engineering
How to Connect an MCP Client to Your Server (HTTP, Auth, and Gateways)
Connect Cursor, Claude Desktop, or your own client to an MCP server: pick transport (stdio vs HTTP/SSE), set the endpoint, pass auth headers, and optionally route through an MCP gateway for one stable URL.
MCP Trail Team
Technical Team
MCP Human-in-the-Loop (HITL): Approval Queues for Risky Tool Calls
Human-in-the-loop for MCP means pausing sensitive tools/call requests until an approver reviews arguments and allows or denies—audit-backed, queue-based, and aligned with compliance-heavy workflows.
MCP Trail Team
Product Team
What Is an MCP Firewall? Gateways, Stable URLs, and Revocable Keys
An MCP firewall (or MCP gateway) sits between AI clients and your MCP servers: one stable URL, Bearer-scoped access, tool policies, secret scanning, rate limits, and audit—without exposing upstream addresses.
MCP Trail Team
Security Team
Argument-level attacks on MCP: when the tool name is allowed but the payload is not
Argument-level MCP attacks: malicious JSON in tools/call while the tool name stays allowlisted. MCP Trail Guardian enforces JSON bounds, DLP on args and results, custom rules, HITL, and catalog policies—see patterns for Slack, Jira, paths, and JSON bombs.
MCP Trail Team
Security
Dangerous MCP tool sequences: real patterns that bypass ‘allowed tool’ lists
Multi-step MCP abuse: each tools/call looks allowed, the chain is not. MCP Trail Guardian tool-sequence policies, risk scoring, HITL, DLP, and catalog allowlists break export→delete, payee→pay, and clone→exfil patterns at the gateway.
MCP Trail Team
Security
How MCP Trail Guardian maps MCP threats to real controls
MCP firewall / security gateway deep dive: SSRF-hardened egress, Bearer auth, catalog allowlists for tools/resources/prompts, DLP, tool sequencing, HITL, JSON argument limits, rate limits, and audit—how Guardian maps each MCP threat class to a real control.
MCP Trail Team
Security
MCP Cost, Token & Performance Optimization: One Playbook
Cut MCP spend end to end: Guardian Smart JSON trim, HTML/CSS strip, identical-call cache (TTL up to 7d) and an optional summarizer, plus budgets, latency and throughput controls that survive production.
MCP Trail Team
Platform
MCP Token Tracking: What to Log and How to Use It
See LLM token use in context: tie usage to MCP servers, clients, and tool rounds. Practical fields to capture, privacy guardrails, and how gateways help you answer who spent what.
MCP Trail Team
Platform
MCP vs Function Calling: Why the Industry is Switching in 2026
Function calling still works. MCP adds discovery, shared context, and a place to hang auth—here is why teams bother switching in 2026.
MCP Trail Team
Technical Team
MCP Access Control & RBAC: Who Can Run Which Tools
Access control for MCP: per-server Bearer tokens, least-privilege tool allowlists, org and per-member scoping, and revocation as the lever that decides who reaches which server and tool.
MCP Trail Team
Security Team
How to Create a Custom MCP Server: Developer Guide
Learn how to build your own MCP server from scratch. Step-by-step guide for creating custom integrations with any tool or service.
MCP Trail Team
Development Team
MCP Security Best Practices: The Complete Checklist (2026)
The canonical, prioritized checklist of MCP security best practices: deny-by-default tool allowlists, argument validation, human approval, DLP, rate and size limits, audit logging, OAuth, session risk scoring, and fail-closed enforcement—each linked to a deep-dive.
MCP Trail Team
Security Team
MCP vs Traditional APIs: Understanding the Differences
Compare Model Context Protocol (MCP) with traditional REST APIs. Learn when to use each approach and how they complement each other in modern AI infrastructure.
MCP Trail Team
Technical Team
MCP vs Webhooks: Pull vs Push, and Why You Often Need Both
Webhooks push event notifications one way; MCP lets an AI agent pull and call governed tools on demand. Compare direction, delivery, and auth—and see how they combine.
MCP Trail Team
Technical Team
Multi-Server MCP Infrastructure: Scale to 50+ Servers and Manage Them at Scale
How to run, route, and govern many MCP servers at once: architecture patterns, a scaling roadmap to 50+ servers, per-server rate limits and budgets, and the management controls you need before an audit team starts asking questions.
MCP Trail Team
Infrastructure Team
Top 10 MCP Servers in 2026: Complete Guide
Discover the most popular MCP servers for AI integration. From Jira to GitHub, Slack to databases—find the right MCP server for your needs.
MCP Trail Team
Research Team
Data retention and compliance gaps
Logs as evidence for regulators and post-incident review—what to retain, what MCP Trail logs on the Guardian path, and what remains your process.
MCP Trail Team
Security
Supply chain and compromised MCP servers
Malicious or hijacked tool definitions undermine trust—what a gateway can and cannot fix.
MCP Trail Team
Security
Shadow IT MCP servers
Unapproved gateways with production access spread inconsistent exposure—why centralizing on Guardian reduces drift.
MCP Trail Team
Security
SSRF and internal network access via tools
MCP SSRF risk when tools fetch URLs or reach internal IPs. MCP Trail Guardian validates upstream hosts, blocks private and metadata ranges by default, enforces Bearer auth, DLP, and audit—plus HITL for high-risk tools. Network design still matters.
MCP Trail Team
Security
Destructive shell and filesystem patterns
rm -rf, path traversal, and pipe-to-shell patterns show up inside MCP tool arguments—definitions and gateway-side mitigations.
MCP Trail Team
Security
Secret exfiltration via arguments or responses
API keys and tokens can leave through tool arguments or echoed responses—how MCP Trail’s gateway-side DLP fits in.
MCP Trail Team
Security
Prompt injection driving tool abuse
Prompt injection can steer models into malicious MCP tools/call payloads. See how MCP Trail Guardian limits blast radius: DLP on arguments and results, HITL, catalog policies, tool sequencing, rate limits, and audit logs—not a substitute for safe prompting.
MCP Trail Team
Security