---
title: "MCP Gateway for Enterprise Security · Highflame"
description: "An MCP gateway that scans servers before agents load them, keeps credentials out of agents' hands, and checks every tool call against policy before it runs."
doc_version: "0.1.0"
last_updated: "2026-11-19T04:03:04.918Z"
canonical: "https://www.highflame.com/platform/mcp-gateway"
---

▸ CONTROL · MCP GATEWAY

# Connect agents to MCP servers. Control every tool call.

Give coding agents secure access to the tools they need without exposing downstream credentials. Highflame scans MCP servers, enforces tool-level policies, and records every action, so teams can move faster without giving up control.

[Book a demo](/contact.md) [See how it works](#how)

Works with Claude Code, Cursor, GitHub Copilot, Windsurf, Gemini CLI, Codex, VS Code, Claude Desktop, ChatGPT, Okta, Microsoft Entra

[SOC 2 Type II, GDPR, HIPAA](/blog/highflame-achieves-soc2-compliance.md)

The problem

## Every MCP connection introduces a new security risk.

MCP servers connect AI agents to code, data, and internal systems. But developers often configure these connections independently, storing credentials locally and connecting tools that security hasn't reviewed.

Once connected, agents may be able to invoke powerful tools without consistent restrictions or visibility into their actions.

### Untrusted tools

tool poisoning

Malicious instructions hidden in MCP tool descriptions can manipulate how agents behave and what they access.

### Exposed credentials

keys in config files

API keys and service credentials stored in local configuration files can be exposed if an agent or developer environment is compromised.

### Uncontrolled access

no tool-level authorization

Without tool-level authorization, agents may invoke tools or perform actions beyond what their tasks require.

4 CLIs

> Claude Code, Gemini CLI, Cursor CLI and Copilot CLI were all open to remote code execution through one malicious .mcp.json file.

[Adversa TrustFall research, via SecurityWeek, May 2026 ↗](https://www.securityweek.com/ai-coding-agents-could-fuel-next-supply-chain-crisis/)

What you get

## Secure MCP access without slowing developers down.

### Approve servers and tools

Scan MCP servers for tool poisoning, malicious instructions, and unexpected changes. Approve trusted servers and enable only the tools your teams need.

### Keep credentials out of agents

Store downstream credentials securely and let the gateway supply them when needed. Agents connect through Highflame without holding the credentials used to access downstream services.

### Enforce policy on every tool call

Check each request against tool permissions and authorization policies before it reaches the destination. Allow, restrict, or deny actions based on your configured controls.

### Know who did what

Record tool requests, policy decisions, and the agent responsible for each action. When linked to Highflame Identity, trace actions back to the person on whose behalf the agent operates.

▸ FIG. 1 · REQUEST PATH / SIGNAL FLOW SCALE 1:1 · ALL DIMENSIONS IN MILLISECONDS · TOL ±0.05

In the product

## Control MCP access down to the tool.

Per-tool control

### Approve individual tools

Enable only the tools agents need. Disable risky or unnecessary capabilities without blocking an entire MCP server.

![](/logos/HF_Icon_Color.svg)MCP Gateway/github-mcp

Tool definitions

**3** of 5 enabled

-   create\_issueOpen an issue in a repository
    
    Clean
-   create\_pull\_requestOpen a pull request
    
    Clean
-   run\_workflowTrigger a CI workflow
    
    Tool poisoning
-   delete\_repositoryDelete a repository
    
    Clean
-   search\_codeSearch code across repositories
    
    Clean

Scanning

### Detect malicious or changed tools

Inspect MCP servers for tool poisoning, hidden instructions, and changes to previously approved tools. Review security findings before enabling access.

![](/logos/HF_Icon_Color.svg)MCP Gateway/jira-mcp

Scanned 2 min ago

2 findings

**1** high**1** medium

-   High
    
    Tool poisoningOWASP MCP Top 10 Hidden instruction in the create\_ticket description tells the agent to forward attachments.
    
-   Medium
    
    Rug pullOWASP MCP Top 10 export\_issues changed its description after the server was approved.
    

Credentials

### Broker credentials securely

Choose shared credentials, OAuth pass-through, or per-user tokens. The gateway handles downstream authentication so agents don't need direct access to service credentials.

![](/logos/HF_Icon_Color.svg)MCP Gateway/salesforce-mcp

Authentication mode

-   Shared credentialOne internal service credential, held by the gateway
-   OAuth 2.1 pass-throughThe person's own sign-in, passed through to the server
-   Per-user tokenA token for each person, fetched at request time

For the teams rolling agents out

## Give teams MCP access without managing keys.

### Simplify connections

✕Before

Every team copies MCP configs and pastes API keys into them.

✓With Highflame

Publish approved MCP servers centrally instead of asking every developer to manage configurations and API keys.

### Speed up security reviews

✕Before

Each new server waits on a manual security review.

✓With Highflame

Use automated scanning and tool-level findings to make informed approval decisions.

### Roll out policies safely

✕Before

New restrictions break workflows the day they ship.

✓With Highflame

Test restrictions in monitor mode before moving to alerts and enforcement, reducing the risk of disrupting existing workflows.

How it works

## From approved servers to authorized actions.

Step

What happens

1 · Approve

Register MCP servers, inspect their tools, and enable only approved capabilities.

2 · Authenticate

Connect agents through Highflame and broker the appropriate downstream credential for each request.

3 · Authorize

Evaluate tool calls against access policies, with controls to allow, redact, hold for approval, defer, or deny requests.

4 · Record

Capture tool requests, policy decisions, and audit evidence so security teams can investigate who or what initiated an action.

Every call through the gateway carries the agent's own identity from Highflame Identity, built on the open-source ZeroID core, so each decision knows which agent acted and for whom. [See how agents get their identity](/platform/agent-identity.md).

▸ PRODUCTION READY

< 50 ms

added to each tool call

0 keys

downstream credentials ever held by an agent

OWASP MCP Top 10

mapping on every scan finding

Open source

Ramparts MCP scanner

## Frequently asked questions

Something else on your mind? [Talk to our team](/contact.md).

01What is an MCP gateway?

A single point that every agent's MCP tool calls pass through, where servers are approved, credentials are supplied, and each call is checked and recorded. Highflame's MCP Gateway also scans servers before they're admitted and keeps every credential out of agents' hands.

02How is this different from a standalone MCP proxy?

A proxy only routes MCP traffic. Highflame also scans each server before agents can load it, keeps downstream credentials out of agents' hands, lets you turn tools on and off one at a time, decides every call with the same policy as your model and agent-to-agent traffic, and keeps a signed record of each call.

03How is an MCP gateway different from an agent security platform?

An MCP gateway routes and secures tool traffic. Highflame's MCP gateway is one part of a platform that also finds agents, issues their identities, and governs model and agent-to-agent traffic under the same policy.

04Can we keep the gateway we already run?

Yes. Highflame can attach to the LLM and MCP gateway you already operate, or run as your MCP gateway. Deploy as SaaS, in your private cloud, or on-prem.

05What if a server changes after we approve it?

Scans check for rug-pulls, where a server changes its tools or behavior after approval, along with tool poisoning and malicious instructions.

06Can people use their company sign-in to reach MCP servers?

Yes. Highflame supports enterprise-managed MCP authorization: it checks the person's corporate sign-in and issues a token scoped to the one MCP server they're calling, and that token is rejected by any other server.

The rest of the platform

## Works with the rest of Highflame

[Identity-Aware Gateway Stop the actions nobody asked for, before they run.](/platform/agent-gateway.md) [Model Gateway Give teams model access and see what each agent spends.](/platform/model-gateway.md) [Agent Discovery Find every agent and MCP server, including shadow ones.](/platform/agent-discovery.md) [Agent Identity Give every agent an identity that leads back to a person.](/platform/agent-identity.md)

Deploy your way

-   **Managed SaaS**Hosted and run by Highflame
-   **Your private cloud**Deployed in your VPC
-   **On-prem**Self-hosted in your environment

SOC 2 Type II · GDPR · HIPAA

DISCOVERY · IDENTITY · CONTROL · GOVERNANCE

## Secure every MCP connection.

See how Highflame protects credentials, governs tool access, and enforces policy across your MCP environment.

[Book a demo](/contact.md)

## Sitemap

Full site map: [/sitemap.md](https://www.highflame.com/sitemap.md). Machine index: [/sitemap-index.xml](https://www.highflame.com/sitemap-index.xml).
