HEADLESS APPS FOR AGENTS

Apps for your agent.
Data you control.

Add tools without changing your harness.

Install apps your agent can use without a graphical interface. Connect once through MCP, then add apps for your work. Your agent can search the marketplace, ask you to approve an installation and use the app. Apps run on your computer or server.

Use your existing harness or connect a standalone agent. Apps can share results so another agent can continue the work. If you only have a model API, an optional runner can track tasks and ask you questions.

Local-first No required UI Your data

Build a tool when you need one

Your agent can turn repeated work into an app, test it and save it for later. Other agents in the same collection can use the app and its results.

Keep the app private or publish it when you choose. How to build an app.

Install Rhyven

Connect Codex, Claude, Cline, Cursor, Hermes, OpenClaw, or another MCP-compatible agent.

Linux x86-64 and ARM64. Docker is optional. Packages use your system package manager; curl installs for your user.

Installable appsLocal execution
Shared schemaMCP and REST access
Custom appsEngine, scripts or containers

Rhyven Marketplace

Browse installable agent-native apps.

Loading apps…

Open an app to see its features, permissions and installation instructions. Apps run in your environment.

APPS WORKING TOGETHER

Use apps together

An agent can create a task, save a decision and send a message using the same three MCP tools. Each app stores its results in the selected collection, where another connected agent can read them.

one interface / saved resultsEXAMPLE
AGENT CAPABILITIES

Test an improvement before reusing it

An agent can improve how it works by saving a missed failure, adding a check and testing both versions on the same cases. This demo connects three apps: the check improves from 2 of 3 correct outcomes to 3 of 3, with no regressions.

Other agents connected to the same collection can reuse the improved policy and its saved results. Keep the app private for your projects, or publish it as open source for others to install. Read the workflow guide.

45 seconds · silent · edited replay of actual MCP calls on prepared fixtures. The agent changes policy data, not executable app code. Download video · Reproduce it
Read the demo transcript
  1. Find the gap. Preflight Checker requires a README and a config file. Both exist, so it accepts the project even though the config contains broken JSON.
  2. Keep the failure. Failure-to-Regression saves the missed failure and a fixture that must be rejected.
  3. Improve the check. The agent adds a JSON-validity check in policy version 2. The original policy remains available.
  4. Measure the change. Workflow Evaluator compares both policies on the same healthy, missing-README and broken-JSON cases. Correct outcomes increase from 2/3 to 3/3, with no regressed cases.
  5. Hand off the result. A fresh MCP connection retrieves the saved evaluation, fixed failure and improved policy. It reruns the malformed config and rejects it.
  6. Reuse the capability. All three apps use the same interface on Rhyven 0.5.5 without Docker. Project state stays private; reusable app packages can be shared separately. This prepared example is not an autonomous-model benchmark.
CUSTOM APP DEVELOPMENT

Create and publish an app

Define objects and actions in a manifest, run Python or JavaScript, or package code in a container. Rhyven provides validation, local storage and a shared agent interface.

Read the developer guide
APP DETAILS