# Connect your agent



Add the MCP server, check access, and give your agent a concrete task.

 

## 1. Create a private client file



Save the client configuration with your deployment’s API URL and token. Restrict access to the file, for example with `chmod 600 /absolute/path/client.json`.

 

## 2. Add the server to your MCP client



Install the Security.Run package so `security-run-mcp` is on your PATH. For clients that accept an `mcpServers` configuration, use the example and replace the path with your private client file. Restart or reconnect your MCP client.

 

## 3. Verify the connection



Ask the agent to call `security_run_list_tools`, then `security_run_inspect_tool` for the tool it plans to use. The inspection response tells it which commands and inputs are available.

 

## 4. Give it a first task



Try: “Use Security.Run to check this file with my YARA rule. Show the matched rules and the original output.” For small files, the agent can call `security_run_upload` with base64 content; the response returns an artifact ID ready to use. That helper is limited to 1.5 MiB; use [HTTP upload](https://www.security.run/docs/upload-inputs.html) for larger files.

 

The agent then calls `security_run_submit_tool_run`, follows the run, and reads its outputs. Ask it to inspect both the exit code and report, rather than treating an accepted request as a completed check.

## Private client.json

```json
{
  "url": "http://127.0.0.1:8000",
  "token": "REPLACE_WITH_YOUR_PRIVATE_TOKEN"
}
```

## MCP client settings

```json
{
  "mcpServers": {
    "security-run": {
      "command": "security-run-mcp",
      "args": [
        "--atomic-only",
        "--config",
        "/absolute/path/client.json"
      ]
    }
  }
}
```