Unified MCP gateway · for security teams

Your whole security stack,
one endpoint for every agent.

Cyproxy federates your threat-intel and detection tools behind a single MCP URL. Point Claude, Cursor, or any agent at one place — and give it scoped, discoverable, fully-audited access to all of it.

Open the console ↗ Read the method

cyproxy · gateway online

ENDPOINT/mcp

VENDORS2 connected

TOOLS51 federated

MODEdiscovery · 3 meta-tools

CREDENTIALSvaulted

AUDIT● enabled

$ claude mcp add --transport http cyproxy /mcp

cyproxy.local / console

◆CyproxyCyber MCP Platform

Dashboard Integrations Logs & Usage Playbooks Team Settings

API calls · this month39 / 50,000

DashboardSearch ⌘K+ Add Integration

Connect. Investigate. Automate.

One MCP endpoint for all your cybersecurity tools.

Your Cyproxy MCP endpointhttp://localhost:8787/mcp/

Copy URL

Connected servers4/4

Federated tools51

Calls (month)39

ToolStatusAuthTools

Team Cymru4 MCP serversconnectedAPI key15

Liberty91Threat IntelligenceconnectedAPI key36

// the console — federate vendors, manage credentials, watch every call

◇

One link.
Every tool your agent can use.

Many clients, many connectors — one audited endpoint in between.

01

From scattered tools
to one controlled surface.

Three moves to hand every agent your whole stack — without wiring a single integration yourself.

01

Register your connectors

Add each MCP server — Team Cymru, Liberty91, your own — under managed auth; keys go straight to the vault. Cyproxy connects and holds a warm pool, isolating any one failure from the rest.

02

Expose one endpoint

Every tool is re-published, namespaced, under a single streamable-HTTP URL. Paste it into any MCP client. No per-tool glue, no new server to deploy.

03

Let agents discover & act

In discovery mode the agent searches a catalog and loads only what it needs — then runs tools by name, scoped and audited, every call logged to the console.

02

Connect. Secure. Optimize.

A gateway you control, sitting between your agents and the tools they investigate with.

◆ Connect

One URL, any connector

Federate stdio and remote MCP servers alike. Add a connector with one registry entry — its actions appear instantly, namespaced, on your endpoint.

◆ Secure

Managed auth, keys vaulted

Credentials stay in the vault and are injected at call time — the agent never sees one. Every call runs as the connected account, through a single enforcement seam, into an immutable audit log.

◆ Optimize

Context that stays lean

Discovery mode shows a handful of meta-tools, not hundreds. Agents search, then load — a fraction of the tokens, none of the clutter.

03

Connect an agent in one line.

No per-tool integration. One URL hands over the entire federated catalog — discovery keeps the agent's context tight.

↳ Discovery

search · catalog · run

Three meta-tools; load only what's relevant.

↳ Vault

Local credentials

chmod-600, injected per call.

↳ Audit

Every call logged

Who, what, which target.

↳ Resilient

Per-server pool

One bad vendor never sinks the rest.

↳ Protocol

Streamable HTTP, done right

Proper init, stateful sessions, version negotiation.

↳ Native

Built on the MCP SDK

Any compliant client connects out of the box.

# point any MCP client at Cyproxy
$ claude mcp add --transport http cyproxy \
    http://localhost:8787/mcp

# the agent discovers what it needs
search_tools("reputation of an IP")
  ↳ scout__get_ip_details
  ↳ scout__get_ip_foundation
  ↳ asn-enricher__ip_to_asn

run_tool("scout__get_ip_details", { ip })
  ✓ 185.220.101.45 — malicious · C2 · botnet

◇

Raw MCP, or a gateway.

What changes when every connector sits behind one controlled endpoint.

CapabilityRaw MCPCyproxy

Connecting clientsOne server per tool, wired per clientOne URL, every tool

CredentialsHanded to each agentVaulted, injected per call

Tool contextAll actions loaded upfrontDiscovery — load on demand

ReliabilityOne bad server breaks the sessionIsolated per-server pool

AuditPer-tool, if anyOne immutable log

Governance—Scope · risk tiers · approvals

04

Pricing that scales with you.

Every plan is the full gateway. You scale on tool calls, audit retention, and team size.

Freeself-serve

£0

For solo analysts getting started.

+ 25,000 tool calls / mo
+ Limited integrations
+ 7-day audit retention
+ 1 team member
+ Community support

Start free ↗

Researchersolo

£29 / mo

Unlimited integrations for one operator.

+ 150,000 tool calls / mo
+ Unlimited integrations
+ 30-day audit retention
+ 1 team member
+ Email support

Choose Researcher ↗

Propopular

£99 / mo

For a small team running real workloads.

+ 750,000 tool calls / mo
+ Unlimited integrations
+ 90-day audit retention
+ 5 team members
+ Priority support

Choose Pro ↗

Businessscale

£499 / mo

High volume, unlimited team, governance.

+ 3,000,000 tool calls / mo
+ Unlimited integrations
+ 365-day audit retention
+ Unlimited team members
+ Priority support

Contact sales ↗

05

Questions, answered.

01What is Cyproxy?+

A gateway that federates your MCP servers — your connectors — behind one URL. Any agent reaches every tool through a single endpoint, with managed auth, tool discovery, and a full audit trail.

02Which clients can connect?+

Any MCP client — Claude, Cursor, ChatGPT, or your own. You paste one URL; there's no per-tool integration to build and no new server to deploy.

03How is authentication handled?+

Managed auth. Credentials live in a vault and are injected at call time — the agent never sees a key, and every call runs scoped to the connected account.

04What is tool discovery?+

Rather than loading hundreds of actions into context, Cyproxy exposes a short discovery toolset — search · catalog · run. The agent finds the right action, then calls it, for a fraction of the tokens.

05Can I add my own connectors?+

Yes. Any MCP server — local stdio or remote streamable-HTTP — registers with one entry, and its actions appear on your endpoint, namespaced and ready to call.

06Who controls access?+

You do. Pre-approve connectors and actions, scope them per member, and review every call in the audit log — governance without getting in the agent's way.

07Self-hosted or managed?+

Run it locally for free, or use the hosted Team and Enterprise tiers for a shared endpoint, SSO, and centralized governance.

08Is it a compliant MCP server?+

Yes. The endpoint speaks streamable HTTP with a proper initialize handshake, stateful session handling via mcp-session-id, and protocol-version negotiation — built on the official MCP SDK, so any compliant client connects out of the box.

One URL. Every tool. Every call accounted for.

// open the console and connect your first vendor in minutes

Open the console ↗

Your whole security stack,one endpoint for every agent.