VC Deal Sourcing via GitHub

Most VC associates build deal-sourcing routines around Crunchbase alerts, AngelList feeds, warm introductions, and Twitter scraping. GitHub-based sourcing complements all of those by surfacing leading-indicator engineering signals three to six weeks before any of those channels do.

The four-step workflow:

1. Define your universe. Pick the sector clusters that match your thesis (ai-ml, fintech, devtools, infra, climate, etc.). GitDealFlow tracks 15 sectors across ~350+ venture-backed orgs. 2. Compute rolling-window metrics weekly. Commit velocity (14-day window), contributor count, new-repo count. The math is documented in the methodology page; the SSRN preprint formalizes it. 3. Classify the acceleration pattern. Each accelerating org maps to one of four patterns, hiring burst, infrastructure buildout, deploy frequency spike, framework migration. The pattern shapes the outreach angle. 4. Rank and outreach. Top quintile by commit-velocity change is your weekly target list. Pair with confirmed-event sources (Crunchbase, PitchBook) to validate. Outreach should reference the *specific* GitHub activity that triggered the signal, generic "saw your traction" notes fail at this layer.

Don't build the pipeline yourself unless you have to. GitDealFlow runs it weekly for 15 sectors and exposes the output via a free MCP server + JSON / CSV / JSONL APIs. The full panel is one curl: curl https://signals.gitdealflow.com/api/signals.json.

For fully agentic sourcing, wire the MCP server into a Claude or OpenAI-Agents runtime with three composed tools: signal-detection (GitDealFlow MCP), enrichment (Crunchbase / Apollo MCP), and draft-outreach (Gmail / HubSpot MCP). The signal layer catches the breakout; enrichment adds firmographic context; outreach drafts a first-touch note. Human-in-the-loop for the final send.

Reading the four acceleration patterns. Each breakout maps to one of four patterns, and each one suggests a different conversation opener. A hiring burst means the team is expanding ahead of a raise, so a note can acknowledge the new contributors on specific repositories. Infrastructure buildout, auth, observability, and billing repos appearing in volume, means the company is hardening to scale, a natural fit for a growth-stage conversation. A deploy frequency spike signals a product push, and a framework migration signals a deliberate tech-stack bet the founders want to ship before the round. Knowing which pattern triggered the signal lets you write a first touch that references the actual activity instead of a generic traction note.

Why you do not need to be an engineer. The metrics are pre-computed and the rankings are sortable, so you only need to know that a sustained velocity change for a venture-backed org means something is happening. The heavy lifting, sector clustering, org filtering, and rolling-window normalization, is done for you and refreshed weekly. You can read the feed the way you read a leaderboard, then hand the names off to your existing sourcing process.

Fitting it into your existing pipeline. Match incoming breakouts against your CRM on the GitHub org URL or the company website domain; both are exposed in the JSON and most systems accept either as a unique key. That gives you a clean dedupe step before a name ever reaches a partner. The same signal also runs in reverse for portfolio monitoring: a fund-of-funds or an existing investor can track engineering acceleration across a portfolio and get an early read on which companies are pulling away.

The export and code-generation surfaces. Beyond the MCP server, the rankings ship as JSON, CSV, and JSONL, and an OpenAPI 3.1 spec describes every callable route so you can generate a client in whatever language you already use. For fully agentic sourcing, compose the signal tool with an enrichment source such as Crunchbase or Apollo and a drafting tool such as Gmail or HubSpot, and keep a human in the loop for the final send. The signal layer catches the breakout, enrichment adds the firmographic context, and outreach drafts the first note.