Practice mode

Practice mode lets anyone try your tools with fake responses. No credential, no sign-in, and nothing reaches your API. It's a safe way to show what your server can do, and to develop an agent against it.

Turning it on

In MCP Studio, turn on Turn on practice mode and publish. Your server then has a separate address:

https://api.octri.dev/api/v1/public/mcp/YOUR_PROJECT_ID/practice/mcp

You can share that address, or offer it as a practice plugin for Claude Code.

To run it on your own machine, or on a package, a bundle, Docker or a Worker:

npx @octri/mcp --project-id YOUR_PROJECT_ID --practice

Or set OCTRI_PRACTICE=1. Packages built while practice mode was off carry nothing to answer from, so publish again first.

What the responses are

Responses are built from your own documentation: the examples in your docs, and your schemas where there's no example. Good examples in your spec make for a more convincing practice session.

Requests are still built and checked by the real code path, so a call with a bad argument fails as it would for real.

Sessions remember

A session keeps what it created. An agent can create a record and then read it back, so a whole workflow can run from start to finish. Each session is separate, so two people never see each other's records.

Simulating failures

Practice sessions get one extra tool, practice_control.

Action	What it does
`simulate_error`	Makes the next call to a named tool fail
`reset`	Clears what the session created

Simulated failures come from your documented error responses, and they go through the same error guidance as real ones. That lets you see how an agent copes with a failure before one happens.

What's different from the real server

	Real server	Practice
Reaches your API	Yes	No
Sign-in or key	Depends on your settings	None
Approval prompts	As you set them	Skipped
Outage notes and change notices	Yes	Skipped
Rate limit	Your limits	Its own limit

Practice answers can be told apart

Agents are told that responses are simulated, so they don't report a practice result to someone as real.

Practice mode or a dry run?

Both keep your API safe, and they answer different questions.

Mode	What comes back	Use it for
Practice mode	Fake responses, so workflows continue	Demos, and developing against your tools
Dry run (`--dry-run`)	The request that would have been sent	Checking exactly what a call would do

Common uses

From the CLI

octri mcp plugin --practice takes the practice version of your Claude Code plugin. See octri mcp plugin.

Your marketing site

Let prospects try an agent on your API before they sign up.

Onboarding

Let new developers explore without touching real data.

Agent development

Build and test an agent, including how it handles failures.