Skip to main content
Section navigation

Creating Agents

Create and configure AI agents on systemprompt.io. Define agent behaviors, assign skills, and set up triggers.

On this page

This guide walks you through creating a new agent on systemprompt.io, from initial setup to testing with Claude.

Before you start

Make sure you have:

  • An active systemprompt.io account with access to the dashboard.
  • At least one skill created. Agents need skills to be useful. If you have not created a skill yet, see the Skills documentation first.
  • Your MCP server connected to Claude (for testing).

Step 1: Navigate to the Agents section

  1. Log in to the systemprompt.io dashboard.
  2. Click Agents in the left sidebar.
  3. You will see a list of your existing agents. If this is your first agent, the list will be empty.

Step 2: Create a new agent

  1. Click the Create Agent button.
  2. Fill in the basic information:
    • Name: A short, descriptive name for the agent. Choose something that clearly identifies its role, like "Blog Writer" or "Support Assistant."
    • Description: A one or two sentence summary of what this agent does. This helps Claude understand when to use this agent. For example: "Writes blog posts and articles in our company's brand voice."

Keep the name concise and the description specific. Vague descriptions like "general helper" make it harder for the platform to route requests to the right agent.

Step 3: Define the agent's persona

The persona controls how the agent behaves and communicates. This is where you shape the agent's identity.

  1. Write a system prompt that tells the agent who it is and how it should operate. A good system prompt includes:

    • The agent's role and purpose.
    • The tone and style it should use.
    • Any boundaries or constraints (what it should and should not do).
    • Key knowledge or context it should always keep in mind.

  2. Here is an example system prompt for a content writing agent:

You are a content writer for a B2B SaaS company. You write in a professional
but approachable tone. Keep sentences short and clear. Avoid jargon unless
writing for a technical audience. Always include a clear call to action.
Do not make claims about product features without being given specific details.
  1. Keep system prompts focused. A 5-10 line system prompt that clearly defines the role works better than a long document trying to cover every scenario.

Step 4: Assign skills

Skills give your agent specific capabilities. An agent without skills has no defined tasks it can perform.

  1. In the agent configuration, find the Skills section.
  2. Click Add Skill to browse your available skills.
  3. Select the skills that match this agent's purpose. For a content writing agent, you might assign:
    • A "Write Blog Post" skill
    • A "Write Social Media Post" skill
    • A "Edit and Proofread" skill
  4. You can assign multiple skills to a single agent. The agent will use the appropriate skill based on the request.

Choose skills that make sense together. A content writing agent should have writing-related skills. Mixing unrelated skills (like "Write Blog Post" and "Analyze Sales Data") on the same agent creates confusion.

Step 5: Add plugins (optional)

Plugins connect your agent to external tools and services. If your agent needs to do more than generate text, add relevant plugins.

  1. In the agent configuration, find the Plugins section.
  2. Click Add Plugin to see available plugins.
  3. Select plugins that support the agent's workflow. For example:
    • A file management plugin if the agent needs to save or retrieve documents.
    • An API plugin if the agent needs to pull data from external services.

Not every agent needs plugins. Start without them and add plugins later when you identify specific integrations the agent requires.

Step 6: Save and review

  1. Review all the settings you have configured:
    • Name and description are clear and specific.
    • System prompt defines the persona and boundaries.
    • Skills are relevant and complementary.
    • Plugins (if any) support the agent's workflow.
  2. Click Save to create the agent.

Step 7: Test your agent

After creating the agent, test it through Claude to make sure it behaves as expected.

  1. Open Claude Desktop or your MCP-compatible client.
  2. Make sure your systemprompt.io MCP server is connected.
  3. Send a request that matches the agent's capabilities. For a content writing agent, try something like: "Write a short blog post about remote work best practices."
  4. Review the output. Check that:
    • The tone matches what you defined in the system prompt.
    • The agent uses the right skill for the request.
    • The output quality meets your expectations.

Step 8: Iterate and improve

Your first version will rarely be perfect. Improve the agent based on testing:

  1. Refine the system prompt. If the tone is off, adjust the persona instructions. If the agent goes out of scope, add clearer boundaries.
  2. Adjust skills. If the agent is missing a capability, create and assign a new skill. If a skill produces poor results, update the skill's instructions.
  3. Test edge cases. Try requests that are slightly outside the agent's intended scope to see how it handles them. Update the system prompt to address any issues.
  4. Add more skills over time. As you identify new tasks the agent should handle, build and assign additional skills.

Example: Creating a customer support agent

Here is a complete example to illustrate the process:

  1. Name: "Support Assistant"
  2. Description: "Answers customer questions about our product features and pricing. Escalates complex technical issues."
  3. System prompt:
You are a customer support assistant. Answer questions clearly and helpfully.
Use a friendly, professional tone. If you do not know the answer, say so honestly
and suggest the customer contact support@example.com. Never guess at pricing
or make promises about upcoming features.
  1. Skills assigned:
    • "Answer Product Questions" -- responds to feature and usage questions.
    • "Explain Pricing" -- walks customers through pricing tiers.
    • "Draft Support Response" -- writes email replies for common issues.
  2. Plugins: None initially. Add a knowledge base plugin later if needed.

Troubleshooting

Agent does not appear in Claude. Make sure your MCP server is connected and running. Restart the MCP server if needed. Verify the agent is saved and not in draft state.

Agent responds with wrong tone or style. Review and update the system prompt. Be more specific about the tone you want. Add examples of good and bad responses if needed.

Agent uses the wrong skill. Check that your skill descriptions are distinct and specific. If two skills overlap in purpose, Claude may pick the wrong one. Clarify the boundaries in each skill's description.

Agent does not use any skills. Verify that skills are assigned to the agent in the dashboard. Check that the skills are published and active.

Next steps

  • Agent Configuration -- Full reference for all agent settings.
  • Skills -- Create more skills for your agents.
  • Plugins -- Extend your agents with external integrations.