Menu

Umbod for platform teams

Manage agent access.
On your own infrastructure.

Connect internal APIs and MCP servers, manage access by group, and run Umbod on your Kubernetes cluster. Keep your connectors and access rules independent of your choice of agent vendor.

Get in touch Self-hosting docs

Early beta for evaluation. See the current limits.

01 / CONNECTION MODEL
Your choice of agent
Claude CodeCodexCursorHermes
Your Kubernetes cluster
umbod Connectors & group permissions
Internal APIsMCP servers
Self-hostedMCP-compatibleIndependent of your agent vendor Explore the architecture

See how it works

From connected tools
to controlled access.

A quick look at connecting tools, granting group access, and choosing your agent in Umbod.

Kubernetes edition · Platform overview Video by community member Sewar Sidou. Thanks, Sewar! Connect tools. Control access. Choose your agent. 22 sec

Why Umbod

One place to manage
who can use which tools.

Direct connections work well to get started. As more teams and agents need access, connector configuration and permissions become harder to keep consistent. Umbod gives platform teams a shared place to manage them.

01 / CONNECT

Configure connectors once.

Connect an internal API through OpenAPI, or add an MCP server. Configure and publish its tools from the administration interface.

02 / CONTROL

Grant access by group.

Choose which tools each group can use. Manage those permissions centrally as teams and their needs change.

03 / KEEP YOUR OPTIONS

Change agents. Keep your setup.

Your connectors and group permissions live in Umbod. Teams can switch compatible MCP clients without moving that configuration into another agent vendor’s platform.

Try it yourself

Start with a
local cluster.

Use kind to try the Kubernetes setup on your machine. The commands below install the beta and open the administration interface.

You’ll need Docker, kind, kubectl, and Helm 3.

  1. Create a cluster

    Terminal
    kind create cluster --name umbod --wait 2m

  2. Install Umbod

    0.0.1-beta.6 chart
    Terminal
    helm upgrade --install umbod \
  oci://ghcr.io/computerlovetech/charts/umbod \
  --version 0.0.1-beta.6 \
  --kube-context kind-umbod \
  --set config.profile=local \
  --set config.authentication.mode=dev \
  --set-string config.publicOrigins.site=http://localhost:3000 \
  --set-string config.publicOrigins.mcp=http://localhost:8011 \
  --wait --timeout 5m

  3. Open the admin UI

    Terminal
    kubectl --context kind-umbod port-forward \
  service/umbod-frontend 3000:3000

    Open localhost:3000. Keep this terminal running while you use Umbod.

  4. Connect a local agent

    Another terminal
    kubectl --context kind-umbod port-forward \
  service/umbod-mcp 8011:8011

    Keep both terminals running. In the admin UI, open MCP setup guide and add http://localhost:8011/mcp as a remote HTTP MCP server in your desktop agent (for example, Claude Code, Codex, or VS Code Copilot). The agent must run on this computer; a hosted agent cannot reach your localhost.

This local setup uses development authentication. For a shared cluster, follow the guide to configure identity, ingress, and storage.

Full installation guide Helm values reference
NEXT

Try a connector you would actually use.

Configure a connector, publish its tools, and grant group access. Then connect your local MCP client. Tell us where you get stuck.

MCP endpoints and ports

Project status

Early beta.
Useful feedback welcome.

Today you can install the Helm chart, manage connectors, and configure group permissions. We’re looking for platform teams to try these workflows and help us understand what’s missing.

The current chart uses a single core replica backed by SQLite. It is an evaluation setup, with no high-availability database architecture yet.