Monitor · Audit · Alert · Discover · Collaborate

The ops layer

MCP was missing.

Teams run 10–20 MCP servers with no visibility into which are slow, which fail silently, or which tools are called most. MCPHub is the control plane they're missing — with multi-tenant workspaces, per-server auth, and a live tool playground.

mcphub.aniruddha.fyi/dashboard
acme-engineering

Total Servers

14

Healthy

12

Tool Calls (24h)

2,847

Avg Latency

23ms

ServerAuthStatusLatencyCalls (24h)
filesystem-mcphealthy2ms1,247
github-mcphealthy18ms891
postgres-mcpwarning124ms432
slack-mcphealthy34ms203

Servers Supported

unlimited

100%

Tool Calls Captured

via transparent proxy

3

Access Levels

member · admin · owner

<1s

Real-time Latency

WebSocket push

The Problem

MCP went from zero to ubiquitous in 8 months.

The protocol is everywhere. The tooling layer doesn't exist yet. Teams manage their MCP servers the same way they managed servers before Kubernetes — manually, with prayer.

Zero visibility

No way to know which servers are slow, which fail silently, or which tools are called most.

Silent failures

Servers degrade. Latency spikes. Error rates climb. Nobody knows until users start complaining.

Ad hoc management

No registry. No audit trail. No cost visibility. Just scattered configs and tribal knowledge.

Agents

Six agents. One control plane.

Each agent handles a specific layer of the ops stack. Together they give you complete coverage of your MCP infrastructure — from discovery to alerting to interactive testing.

Server Registry

Catalog of all MCP servers with metadata, version, owner, and per-server auth credentials.

Health Prober

On-demand pings measuring latency, error rate, and availability across every registered server.

Tool Call Logger

Intercepts every tool invocation through the transparent proxy — duration, output size, caller identity.

Usage Analytics

Which tools get called most, by which agents, at what cost. Pre-aggregated hourly for instant queries.

Alert System

Configurable rules fire when a server degrades or goes offline. Slack webhook + generic HTTP delivery.

Tool Playground

Dynamically discover every tool a server exposes and invoke them interactively — all logged to the audit trail.

Multi-Tenant

Built for teams.

Full workspace isolation with granular access levels. Invite teammates, manage API keys, and keep every team's data strictly scoped — all the way down to the Redis cache prefix.

L1

Member

  • View dashboard & servers
  • Browse tool calls & analytics
  • Read alert history
L2

Admin / Owner

  • Register & edit servers
  • Invoke tools in Playground
  • Manage members & invites
  • Create API keys
API

API Access

  • Workspace-scoped API keys
  • Bearer token or X-API-Key header
  • Same permissions as issuing role
  • Revocable at any time

Per-server auth configuration

Each MCP server can be configured with its own auth credentials — bearer token, API key header, or HTTP basic auth. Credentials are masked in responses and forwarded transparently by the proxy and health prober. No changes to existing server infrastructure required.

Bearer tokenAPI key headerHTTP basic auth
Tool Playground

Discover and test every tool — live.

MCPHub dynamically fetches the full tool manifest from any registered server via JSON-RPC tools/list, renders each tool's JSON Schema as an interactive form, and lets admins invoke them — all logged to the audit trail as mcphub-playground.

Dynamic discovery

Fetches the live tool list from any server on demand. Results are Redis-cached for 5 minutes with a manual refresh option.

Interactive invocation

JSON Schema input fields are rendered as a form automatically. Raw JSON mode available for complex nested schemas.

Full audit trail

Every playground invocation is logged to tool_calls with caller_agent="mcphub-playground" — visible in the audit log immediately.

Tool Playground request flow

ToolsTab

GET /servers/{id}/tools

MCPHub Backend

Redis-cached 5min

MCP Server

tools/list · tools/call

Architecture

How it works.

MCPHub sits as a transparent reverse proxy in front of your MCP servers. Zero changes required to existing infrastructure.

01

Create your workspace

Sign up and get a personal workspace instantly. Invite team members with granular roles — member, admin, or owner.

02

Register your servers

Add MCP server endpoints with metadata, ownership, and optional per-server auth (bearer token, API key, or HTTP basic).

03

Route through the proxy

Point your MCP clients to MCPHub's transparent proxy. Zero changes to existing servers. Every call is logged automatically.

04

Monitor, test, and alert

Live health status and tool call volume on the dashboard. Explore tools in the playground. Get alerted when servers degrade.

Request flow

MCP Client

Claude / Agent

MCPHub Proxy

/proxy/{id}/mcp

MCP Server

filesystem / github

PostgreSQL

audit log + analytics

Redis

pub/sub · streams · cache

WebSocket

real-time push

JWT + API Keys

workspace-scoped auth

Get Started

Start monitoring your

MCP layer today.

The gap is wide open. Be the team with visibility before everyone else catches up.