Pixra

Install the PIXRA package from PyPI into your Python environment. This pulls in the core agent harness, verifier layer, and MCP integration modules.

pip install pixra

Export your OpenAI API key as an environment variable. PIXRA uses this for all LLM calls including the Verifier Agent and Editor Agent stages. Set UPSONIC_TELEMETRY to False if you want to opt out of anonymous usage tracking.

export OPENAI_API_KEY=sk-your-openai-key-here
export UPSONIC_TELEMETRY=False

Import Task and Agent from the upsonic module that PIXRA builds on. Create a Task with your prompt, instantiate an Agent, and call print_do to run the full verified pipeline.

from upsonic import Task, Agent

task = Task("Summarise the top 5 Python web frameworks in 2025")
agent = Agent("Researcher")
agent.print_do(task)

For production use, define a Pydantic response model so the Verifier Agent can check that all required fields are present and correctly typed before returning results.

from pydantic import BaseModel
from upsonic import Task, Agent

class FrameworkSummary(BaseModel):
    name: str
    stars: int
    description: str

task = Task(
    "List the top 3 Python web frameworks",
    response_format=list[FrameworkSummary]
)
agent = Agent("Researcher")
result = agent.do(task)
print(result)

PIXRA natively supports MCP servers as tool providers. Pass MCP server configurations to the agent so it can call external tools during task execution without custom wrappers.

from upsonic import Agent

agent = Agent(
    "Assistant",
    mcp_servers=["your-mcp-server-command"]
)
agent.print_do(task)

To expose a PIXRA-backed agent as an MCP server for Claude Desktop, add the entry to your claude_desktop_config.json. Use uvx to run it without a manual pip install step.

{
  "mcpServers": {
    "pixra": {
      "command": "uvx",
      "args": ["pixra"],
      "env": {
        "OPENAI_API_KEY": "sk-your-openai-key-here",
        "UPSONIC_TELEMETRY": "False"
      }
    }
  }
}