dm-sparkSDK 0.1.1

No project description provided

Spark SDK

Spark SDK is a lightweight Python client for the DeepModel Spark API. It gives your application programmatic access to MCP tools and worker agents — allowing you to list available tools and invoke them directly, without any LLM inference overhead.

---

Features

• List all available MCP tools and worker agents with connection status and payload schemas
• Directly invoke any MCP tool or worker agent with structured arguments
• Clean, async-first API built on httpx
• Consistent error handling — errors are returned in the response, never raised unexpectedly

---

Requirements

• Python >= 3.8
• httpx >= 0.24.0

---

Installation

pip install dm-sparkSDK

---

Configuration

When initializing the client, you must provide the following:

Parameter	Type	Required	Description
api_key	str	✅ Yes	Your DeepModel API key for authentication (x-api-key)
api_url	str	✅ Yes	The base URL of your DeepModel Spark API instance

import dm-sparkSDK

client = dm-sparkSDK.Client(
    api_key="your_api_key",
    api_url="https://your-api-url.com"
)

You can store these values as environment variables and load them with os.environ or a library like python-dotenv:

import os
import dm-sparkSDK

client = dm-sparkSDK.Client(
    api_key=os.environ["SPARK_API_KEY"],
    api_url=os.environ["SPARK_API_URL"]
)

Recommended Environment Variables

Variable	Description
SPARK_API_KEY	Your DeepModel API key
SPARK_API_URL	The base URL of the Spark API instance

---

Constants (Enums)

The SDK exposes target type constants used when invoking tools:

Constant	Value	Description
dm-sparkSDK.MCP	"mcp"	Target an MCP tool
dm-sparkSDK.AGENT	"agent"	Target a worker agent

import dm-sparkSDK

dm-sparkSDK.MCP    # "mcp"
dm-sparkSDK.AGENT  # "agent"

---

Available Methods

client.get_unified_tools()

Retrieves all MCP tools and worker agents accessible to the authenticated user, including their connection status and payload schemas.

result = await client.get_unified_tools()

Returns: dict

Field	Type	Description
success	bool	Whether the request was successful
tools	dict	Contains mcp_tools (grouped by server) and worker_agents
status_code	int	HTTP status code of the response
detail	str / null	Error message if the request failed

Example response:

{
    "success": true,
    "tools": {
        "mcp_tools": {
            "github": [
                {
                    "tool_name": "create_issue",
                    "is_configured": true,
                    "connection_id": "uuid-here",
                    "description": "Creates a GitHub issue",
                    "payload_schema": {},
                    "icon_url": "https://..."
                }
            ]
        },
        "worker_agents": [
            {
                "tool_name": "My Agent",
                "is_configured": true,
                "connection_id": "uuid-here",
                "description": "A custom worker agent",
                "payload_schema": {},
                "agent_avatar_url": "https://..."
            }
        ]
    },
    "status_code": 200,
    "detail": null
}

---

client.invoke_tool()

Directly invokes an MCP tool or worker agent with the provided arguments. No LLM inference is performed — arguments are passed as-is.

result = await client.invoke_tool(
    connection_id="uuid-here",
    target_type=dm-sparkSDK.MCP,
    tool_or_agent_name="create_issue",
    arguments={"title": "Bug report", "body": "Something went wrong"},
    query="create a github issue"   # optional
)

Parameters:

Parameter	Type	Required	Description
connection_id	str / UUID	✅ Yes	The connection ID of the tool or agent (from get_unified_tools)
target_type	str	✅ Yes	dm-sparkSDK.MCP or dm-sparkSDK.AGENT
tool_or_agent_name	str	✅ Yes	The exact name of the tool or agent to invoke
arguments	dict	✅ Yes	Key-value arguments matching the tool's payload_schema
query	str	❌ No	Optional natural language context for agent invocations

Returns: dict

Field	Type	Description
success	bool	Whether the invocation was successful
tool_or_agent_name	str	The name of the tool or agent invoked
connection_id	str	The connection ID used
result	any	The output returned by the tool or agent
status_code	int	HTTP status code of the response
detail	str / null	Error message if the invocation failed

---

Full Example

import asyncio
import os
import dm-sparkSDK

async def main():
    client = dm-sparkSDK.Client(
        api_key=os.environ["SPARK_API_KEY"],
        api_url=os.environ["SPARK_API_URL"]
    )

    # Step 1: Get all available tools
    tools_result = await client.get_unified_tools()

    if not tools_result.get("success"):
        print(f"Failed to fetch tools: {tools_result.get('detail')}")
        return

    tools = tools_result["tools"]
    print("MCP Tools:", tools.get("mcp_tools"))
    print("Worker Agents:", tools.get("worker_agents"))

    # Step 2: Invoke an MCP tool
    invoke_result = await client.invoke_tool(
        connection_id="your-connection-id",
        target_type=dm-sparkSDK.MCP,
        tool_or_agent_name="create_issue",
        arguments={
            "title": "Bug report",
            "body": "Something went wrong"
        },
        query="create a github issue"
    )

    if not invoke_result.get("success"):
        print(f"Invocation failed [{invoke_result.get('status_code')}]: {invoke_result.get('detail')}")
    else:
        print("Result:", invoke_result.get("result"))

    # Step 3: Invoke a worker agent
    agent_result = await client.invoke_tool(
        connection_id="your-agent-id",
        target_type=dm-sparkSDK.AGENT,
        tool_or_agent_name="My Agent",
        arguments={"input": "Hello, agent!"}
    )
    print("Agent Result:", agent_result.get("result"))

asyncio.run(main())

---

Error Handling

All errors are returned within the response dict — no exceptions are raised unexpectedly.

result = await client.get_unified_tools()

if not result.get("success"):
    print(f"Error {result.get('status_code')}: {result.get('detail')}")

Error Type	status_code	detail Example
HTTP Error	e.g. 401	"No detail provided"
Request Error	400	"Request error: Connection refused"
Unexpected Error	500	"An unexpected error occurred: ..."

---

License

MIT License