Phase 3E — Multi-Agent Convergence

Multi-Agent AI
Research Platform

Discover, analyze, and synthesize research papers using collaborative AI agents. MACP v2.0 enables Gemini, Claude, GPT-4o, and Grok to converge their analyses into a unified knowledge base on GitHub.

Try Live Demo View Source Quick Start Guide
6
AI Agents
12,800+ Papers Searchable
8
WebMCP Tools
v2.0
MACP Schema

Research Superpowers

Everything you need to accelerate your AI research workflow, powered by multi-agent collaboration.

🔍

Multi-Source Paper Discovery

Search across arXiv, HuggingFace Daily Papers, and Semantic Scholar simultaneously. Find the most relevant papers with intelligent ranking.

🧠

Multi-Agent Analysis

Get research papers analyzed by Gemini, Claude, GPT-4o, and Grok. Each agent brings unique strengths — compare their perspectives side by side.

📊

Consensus Synthesis

When multiple agents analyze the same paper, MACP generates a consensus report showing agreement scores, convergence points, and divergence areas.

💾

GitHub-First Persistence

Every paper, analysis, and note is stored as a structured JSON file in your GitHub repository. Your research journey is version-controlled and permanent.

🔑

Bring Your Own Key (BYOK)

Use Gemini for free, or bring your own API keys for Claude, OpenAI, or Grok. Validate and apply keys with one click. Your keys are never stored, logged, or transmitted to us — verified by code audit.

🕸️

Knowledge Graph

Visualize citation networks and topic relationships across your research library. Discover connections between papers that you might have missed.

🔒 BYOK PRIVACY GUARANTEE

Your API Keys Stay Yours

When you use Bring Your Own Key (BYOK), your API key is used only for the duration of your request and is immediately discarded. Keys exist only in memory during the API call — they are never stored in our database, never written to GitHub, never logged, and never included in API responses or provenance tracking. This has been verified through a comprehensive code audit.

✗
Not stored in DB
✗
Not written to GitHub
✗
Not in logs
✓
Request-only memory

Agent Ecosystem

Six AI agents with complementary strengths, all contributing to your research through the MACP v2.0 protocol.

Google Gemini
gemini-2.5-flash
Free
Anthropic Claude
claude-sonnet-4-5
BYOK
OpenAI GPT-4o
gpt-4o
BYOK
xAI Grok
grok-3
BYOK
Perplexity Sonar
sonar-pro
Deep Research
Manus AI
CSO R — Orchestrator
FLYWHEEL

MACP v2.0 Directory Standard

A self-describing schema that any AI agent can read, understand, and contribute to.

# Every research journey is stored as structured data on GitHub

.macp/
├── schema.json                                    ← Self-describing spec
├── manifest.json                                  ← Master index
├── papers/
│   └── {arxiv_id}.json                           ← One per paper
├── analyses/
│   └── {arxiv_id}/
│       ├── gemini_20260224.json                  ← Gemini's analysis
│       ├── anthropic_20260224.json               ← Claude's analysis
│       └── consensus.json                        ← Multi-agent synthesis
├── agents/
│   ├── gemini.json                               ← Agent registry
│   ├── anthropic.json
│   ├── openai.json
│   ├── grok.json
│   ├── perplexity.json
│   └── manus.json
├── notes/
│   └── note_{uuid}.md                            ← Research notes
├── graph/
│   └── knowledge-graph.json                      ← Citation network
└── handoffs/
    └── {date}_{author}_{topic}.md                ← Team alignment docs

Development Roadmap

Built by the FLYWHEEL TEAM — a multi-agent collaboration between human orchestrator, Manus AI, and Claude Code.

Phase 1-2 · Jan 2026

Foundation & CLI Tools

MACP methodology, paper fetcher (arXiv, Semantic Scholar, HuggingFace), citation tracker, knowledge graph generator. 55 research papers collected.

Phase 3A-3C · Feb 2026

Web Platform & Production

React + FastAPI web UI, 8 WebMCP tools, GitHub OAuth, GCP Cloud Run deployment, multi-provider LLM support (Gemini/Claude/GPT-4o/Grok), BYOK.

Phase 3D · Feb 2026

GitHub-First Persistence

Save pipeline fix, My Library, BYOK Validate & Apply UX, .gitignore fix, GitHub dual-write, MACP v2.0 directory initialization.

Phase 3E · Feb 2026

Multi-Agent Convergence

MACP v2.0 schema specification, deep PDF analysis (PyMuPDF), per-agent analysis files, agent registry (6 agents), consensus synthesis (40/30/30 scoring), Perplexity deep research API integration, 13 MCP endpoints.

Phase 3F · Feb 2026

Deployment & UI Polish

Cloud Run revision 00022, Agent Registry UI (card grid, cost tiers, capability chips), Consensus Comparison UI, Deep Analysis View, resizable detail panel, 10/10 Dependabot PRs closed, CI all green, 0 code scanning alerts.

Phase 4 · Q1-Q2 2026 · CURRENT

Knowledge Intelligence

Interactive knowledge graph visualization, citation network analysis, cross-paper methodology comparison, automated literature review generation, research templates.

Phase 5 · Q2-Q3 2026

Community & Ecosystem

Shared research collections, collaborative annotations, public knowledge graphs, Zenodo integration for defensive publications.

Start Researching with AI Agents

MACP Research Assistant is fully open source. Try the live demo, explore the codebase, or contribute to the project.

Launch App Join Discussion