MCP Server

Connect any MCP client to PT-Edge for live AI ecosystem intelligence.

220K+ GitHub repos across 30 AI domains — quality scores, star velocity, downloads, lifecycle stages, and semantic search. Updated daily.

Quick Start — Claude Desktop

Add this to your claude_desktop_config.json:

{
  "mcpServers": {
    "pt-edge": {
      "type": "streamableHttp",
      "url": "https://mcp.phasetransitions.ai/mcp/stream"
    }
  }
}

With authentication (higher rate limits):

{
  "mcpServers": {
    "pt-edge": {
      "type": "streamableHttp",
      "url": "https://mcp.phasetransitions.ai/mcp/stream?token=pte_YOUR_KEY"
    }
  }
}

Restart Claude Desktop after saving. PT-Edge will appear in the server list.

Get a free API key →

Other Clients

Claude.ai (web)

Go to Settings → MCP Servers → Add. Endpoint:

https://mcp.phasetransitions.ai/mcp

Cursor / Windsurf

Use the streamable HTTP endpoint:

https://mcp.phasetransitions.ai/mcp/stream

Python SDK

from mcp import ClientSession
from mcp.client.streamable_http import streamablehttp_client

async with streamablehttp_client("https://mcp.phasetransitions.ai/mcp/stream") as (r, w, _):
    async with ClientSession(r, w) as session:
        await session.initialize()
        result = await session.call_tool("get_status", {})

What You Get

PT-Edge tracks 220,000+ open-source AI repos across 30 domains — from MCP servers and agents to RAG, embeddings, voice AI, and ML frameworks. Each repo is scored 0–100 on maintenance, adoption, maturity, and community signals. Data flows in daily from GitHub, PyPI, npm, HuggingFace, and Hacker News.

Sample Conversation

What are the best autonomous coding agents?

find_ai_tool("autonomous coding agent", domain="agents")

Tool response
Name                  Stars    Score  Description
openai/codex          15,842   91     Lightweight coding agent that runs in the terminal
Pythagora-io/gpt-pilot 32,410  87     AI developer that builds apps from spec
OpenDevin/OpenDevin    28,193  85     Open-source autonomous software engineer

The top autonomous coding agents right now are Codex (91 quality score, 15.8K stars), GPT Pilot (87, 32.4K stars), and OpenDevin (85, 28.2K stars). Codex is the newest but has the highest quality score due to strong maintenance and adoption signals.

Compare the top two

query("SELECT full_name, stars, quality_score, downloads_monthly FROM mv_mcp_quality WHERE full_name IN ('openai/codex', 'Pythagora-io/gpt-pilot')")

Tool response
full_name              stars   quality_score  downloads_monthly
openai/codex           15,842  91             284,000
Pythagora-io/gpt-pilot 32,410  87             51,200

Codex leads on quality score (91 vs 87) and monthly downloads (284K vs 51K). GPT Pilot has more stars (32K vs 16K) and a longer track record. Codex is the stronger pick if you want active maintenance; GPT Pilot if you want battle-tested community support.

What's trending in that space this week?

query("SELECT full_name, stars, stars_7d_delta, quality_score FROM mv_project_summary WHERE domain = 'ai-coding' ORDER BY stars_7d_delta DESC LIMIT 5")

Tool response
full_name                stars   stars_7d_delta  quality_score
openai/codex             15,842  +2,340          91
block/goose              12,847  +890            82
cline/cline              28,441  +670            88
Codium-ai/pr-agent       7,921   +410            79
aider-ai/aider           31,205  +380            90

Codex is dominating this week with +2,340 stars. Goose and Cline are also surging. The AI coding space is one of the most active domains right now.

Core Tools (8)

A slim tool surface — discover the schema, compose SQL, get answers.

get_status()

Start here. Returns table count, repo count, domains, freshness, and guidance.

Tables: 142 | AI repos: 220,457 | Domains: 30

list_tables()

All database tables with row counts.

describe_table(table_name)

Column names, types, and nullability.

describe_table("ai_repos")

search_tables(keyword)

Find tables by keyword in table or column names.

search_tables("embedding")

query(sql)

Run any SELECT. Read-only, 5s timeout, 1000 row limit.

query("SELECT full_name, stars FROM ai_repos WHERE domain = 'mcp' ORDER BY stars DESC LIMIT 5")

list_workflows()

Pre-built SQL templates for common questions. Adapt and run via query().

find_ai_tool(query, domain?, limit?)

Semantic search across 220K+ repos. Describe what you need in plain English.

find_ai_tool("vector database for production", domain="vector-db")

submit_feedback(topic, text, category?)

Report bugs, request features, or share observations about the data.

Recommended Workflow

  1. get_status() — see what's available
  2. list_tables() — browse the schema
  3. describe_table('ai_repos') — see columns
  4. list_workflows() — get recipe templates
  5. query('SELECT ...') — get your answer

For semantic similarity search, use find_ai_tool() instead of SQL.

Authentication

No auth required for basic access (rate limited). Get a free key at /api/docs#get-a-key for higher limits. Pass as query param or header:

?token=pte_YOUR_KEY
Authorization: Bearer pte_YOUR_KEY

Also Available