Managing agents from AI (MCP)

Once you've connected Contextium to your AI tool (Claude Desktop, VS Code, or Cursor), you can manage your agents entirely through conversation. You don't need to open the Contextium dashboard at all. The same MCP connection that lets your AI browse your workspace also gives it tools to create, update, and run agents.

If you haven't set up the MCP connection yet, see Using MCP.

What your AI can do with agents

Your connected AI assistant has built-in tools for working with agents:

• Create an agent — name it, write a system prompt, and set a description
• List agents — see all agents in a workspace
• Update an agent — change the name, description, or system prompt
• Delete an agent — remove an agent from a workspace
• Dispatch a job — ask an agent to carry out a task and return the result
• Check job status — poll a running job to see if it's finished and retrieve the output

You don't need to know the tool names or any syntax. Just describe what you want in plain English and your AI handles the rest.

Creating and managing agents via conversation

Here are some things you can say to your AI assistant:

• "Create an agent called 'Release Notes Writer' with a system prompt that writes concise release notes from a list of commit messages."
• "List all agents in my workspace."
• "Update the 'Support Agent' — add a line to the system prompt saying it should always ask for the user's account ID before troubleshooting."
• "Delete the 'Test Agent' from my workspace."

Your AI will confirm what it's about to do and ask for any missing details before taking action. It won't delete or overwrite anything without your input.

Dispatching a job to an agent

Dispatching a job means asking one of your Contextium agents to carry out a task and return a result. This is useful when you want to run a task in the background — for example, generating a document, summarising some content, or reviewing a file — and get back a finished output.

To dispatch a job, describe the task in plain English:

"Ask the Technical Writer agent to summarise this changelog and turn it into a user-facing release announcement."

"Use the Code Review agent to review this function and flag any issues."

"Have the Support Agent draft a response to this customer message."

Your AI will call dispatch_agent_job behind the scenes, sending your request to the right agent along with any content you've provided. It then waits for the result or lets you know the job has been queued.

Checking on a running job

For longer tasks, a job runs in the background and you can check back on it at any time. Just ask:

"What's the status of that release notes job?"

"Has the Support Agent finished drafting that reply?"

Your AI calls get_agent_job to fetch the current status. When the job is complete, it returns the output directly in the conversation so you can review, edit, or use it right away.

When to use the AI vs the dashboard

The AI path is fastest for quick changes — creating an agent on the fly, updating a prompt mid-session, or dispatching one-off tasks. The dashboard is better for visual browsing, managing skills, and working with marketplace listings.

Both paths are fully synced. Anything you create or change via the AI appears immediately in the dashboard, and vice versa.