Run your first local session
In this tutorial, you will start mecatui in a local project and ask Mecatl to
inspect it. mecatui runs a private embedded server, so you do not need to
start a separate service.
Prerequisites
You need:
- macOS or Linux with Homebrew;
- an API key with API access enabled from Anthropic, OpenAI, OpenRouter, or OpenCode; and
- a local project directory that you trust.
The project is your workspace. Mecatl limits its file tools and commands to this directory.
This guided path uses provider API credentials. A ChatGPT Plus or Pro subscription does not provide an OpenAI API key and cannot be used to sign in through this setup. Mecatl's OIDC options authenticate to an operator-configured gateway or remote Mecatl server, not to a consumer provider account. See Provider authentication for the supported paths.
Install Mecatl
Homebrew installs the mecatui client and the mecated server:
brew install stacklok/tap/mecatl
mecatui --version
The version command should print a release tag.
Set up a provider
Use an API key for this tutorial. Run the interactive setup and select your provider:
mecatui providers setup
Enter the API key when prompted. mecatui saves locally managed credentials
without displaying them. Confirm that your provider reports ready to use:
mecatui providers status
For a custom provider or gateway, follow Configure a custom provider. Custom provider IDs use lowercase letters, numbers, and hyphens.
Start mecatui
cd <PROJECT_DIRECTORY>
mecatui
To submit the first prompt at startup while keeping the session interactive,
pass --prompt:
mecatui --prompt "Summarize the failing tests in this repository"
The welcome screen shows your workspace and active model. Keep the default model until you complete the first request.
The header also shows mode default, identifying the active permission mode.
With the default permission policy, read-only tools can run without approval and
actions that change the workspace ask first.
Inspect the project
Enter this request:
Inspect the top-level files and explain what this project does. Cite the files you used.
You should see tool cards as Mecatl reads the workspace, followed by an answer
based on your project. Press ctrl+t on a tool card to inspect its full input
and result.
You now have a local session: the conversation, selected model, workspace,
and tool history that Mecatl keeps together. Exit with ctrl+c twice or
/quit. To continue the newest stored session, run:
mecatui --resume-latest
Try another model
Enter /models to open the models reported by your configured provider. Select
a model and press Enter. Model IDs and capabilities are provider-specific,
so start with a model shown in this inventory. Changing models creates a peer
session and carries over the visible conversation.
If a listed model fails, see Troubleshoot provider and model setup.
Add tools from local MCP servers (optional)
MCP servers give agents tools for working with external services and data. ToolHive is Stacklok's open source runtime for running MCP servers locally.
If ToolHive has MCP servers running in its default group, the embedded server
discovers them at startup. Enter /mcp to inspect the available MCP sources and
tools. Mecatl connects to those servers but does not start them.
Next steps
- Connect to a separate server to move the server out of
the
mecatuiprocess. - Work in the TUI to steer runs, review tools, and approve actions.
- Manage sessions to resume, inspect, and fork conversations.
Troubleshooting
If provider setup or the first model request fails, run
mecatui providers status and use the recovery action shown for that provider.
For custom gateways, verify the provider ID, API flavor, exact model ID, and
gateway model access in
Troubleshoot mecatui.