DeepSeek Harness Plugins: Connect Your Tools and Extend What the Agent Can Do

What Harness plugins are

An AI agent is only as useful as the things it can reach. Harness on its own can read and write files in your chosen folder and run commands there. A plugin adds new tools, actions the agent can choose to use, such as "search the web", "open a pull request", "send a message" or "run this SQL query".

DeepSeek describes Harness plugins as composable. You can turn on several at once, and the agent decides which to use for each step. A single task can read your files, look something up online, update a spreadsheet and post a summary to your team.

How plugins work

Every plugin tool has a name, a plain-language description of what it does, and the inputs it needs. The model reads those descriptions and decides when a tool is the right next step, the same way tool calling works in the DeepSeek API. The plugin carries out the action and returns the result, and the agent carries on from there.

Types of plugins

What's available depends on your Harness version and on what the community has built. These are the most common kinds of plugin and what they make possible:

Developers building something similar into their own app can use the DeepSeek API's search - fetch tools and function calling.

Install and enable a plugin

Combining plugins

The real power comes from using several plugins in one task. Some combinations that work well:

Run any of these automatically with scheduled tasks.

Worked examples

Example 1 - Web plugin: keep a price list current

Example 2 - Git hosting plugin: from issue to pull request

Build your own plugin

Because Harness is open source, you can write plugins for anything your own systems can do, such as an internal API, a company database or a device on your network. Conceptually, a plugin is just a set of tools, and each tool has three parts:

{
  "name": "get_order_status",
  "description": "Look up the current status of a customer order by its ID. Use when the user asks where an order is or whether it has shipped.",
  "parameters": {
    "type": "object",
    "properties": {
      "order_id": { "type": "string", "description": "Order number, e.g. 48213" }
    },
    "required": ["order_id"]
  }
}

import os, requests

def get_order_status(order_id: str) -> dict:
    """Read-only lookup against our internal orders API."""
    if not order_id.isdigit():
        return {"error": "order_id must be numeric"}
    r = requests.get(
        f"https://orders.internal.example/api/orders/{order_id}",
        headers={"Authorization": f"Bearer {os.environ['ORDERS_TOKEN']}"},
        timeout=10,
    )
    if r.status_code == 404:
        return {"error": "order not found"}
    o = r.json()
    # Return only what the model needs, keeping it short and free of secrets
    return {"order_id": order_id, "status": o["status"], "eta": o.get("eta")}

How the model chooses and calls tools is explained in the API guide · tools. A complete tool-calling loop is shown in search - fetch · do-it-yourself.

Plugin design tips

Permissions and safety

General Harness safeguards: permissions - safety.

Troubleshooting

Harness plugins - frequently asked questions