Documentation

Everything you need to build powerful LLM agents with MCP integration using Pulsar Agent framework.

Getting Started

Pulsar Agent is an open-source, evolving LLM agent framework designed to integrate seamlessly with Claude's MCP (Model Context Protocol) server ecosystem.

Installation

In the default configs.json, Pollinations AI (https://pollinations.ai/) is used as the LLM provider. You can refer to the configs-example-1.json to use other OpenAI compatible API and configs-example-2.json to use local deployed Ollama.

git clone https://github.com/aeromechanic000/pulsar-project.git
cd pulsar-project
uv run app.py
  • Add --save-logs to enable logging into ./logs/
  • Navigate to http://localhost:9898 in your browser

CLI Client Mode

Run the standalone agent client:

uv run client.py

Options:

  • --config-path <path>: specify a custom config file (default: configs.json)
  • --save-logs: enable detailed logging of prompts, responses, and actions

Core Concepts

Understanding the fundamental concepts behind Pulsar Agent's MCP-integrated architecture.

Modular MCP Integration

Connects to multiple MCP servers using Claude's protocol. Acts as a standalone MCP client with full tool discovery and execution. Provides uniform access to distributed toolsets from various servers.

Task-Centric Architecture

Uses a TaskManager to manage task contexts and workflows. Developers can create custom task types (e.g., planning, research). Logs and state are persistently tracked per task for traceability.

Memory-Driven Agent Intelligence

Built-in Memory module for long-term context accumulation. Extracts and stores structured information to enhance reasoning and multi-turn interaction. Supports memory operations dynamically invoked by LLM.

Web-based Interaction

A full-featured Flask app provides RESTful APIs for task/memory/tool interactions. Live chat interface powered by WebSocket. Task visualization, memory inspection, and tool usage monitoring.

API Reference

Complete reference for Pulsar Agent's REST API endpoints.

API Endpoints

Flask app provides REST APIs:

  • POST /api/initialize: initialize agent with config
  • POST /api/chat: submit query to agent
  • GET /api/config: fetch current agent + tool status
  • GET /api/tasks: view all tasks
  • GET /api/memory: view memory summary
  • GET /api/tools: list tools from all connected MCP servers

Structured Response

Agent LLM will generate JSON responses with:

  • think: reasoning
  • text: final answer
  • mem_op: memory operation (if needed)
  • tool: tool usage (if needed)
  • finished: whether interaction is complete

Logs and Configuration

Managing your Pulsar Agent setup and logging.

  • Logs are stored in ./logs/ if --save-logs is enabled
  • Agent behavior is controlled via configs.json:
    • Provider selection
    • MCP server addresses
    • Task workflow configuration
    • Memory strategies

Table of Contents