GitHub Commit Velocity Tracker API

For agents and pipelines that need a turn-key GitHub commit velocity tracker API, GitDealFlow's public endpoints return the metric pre-computed across ~350+ venture-backed startup organizations spanning 15 sectors.

What's in the panel. Every tracked startup carries: commitVelocity (total commits in the rolling 14-day window), commitVelocityChange (percentage delta vs. the prior window, the primary ranking signal), contributors (unique contributor count), signalType (one of five classification categories), an estimated stage, a sector slug, and a list of recentRepos for the 30-day repository-creation window.

Access patterns.

```bash # Bulk panel, all sectors, all periods, all startups curl https://signals.gitdealflow.com/api/signals.json

# Single startup lookup curl "https://signals.gitdealflow.com/api/signal?name=Roboflow"

# CSV for spreadsheets / dataframes curl https://signals.gitdealflow.com/api/signals.csv

# MCP server, invokable from Claude / Cursor / Windsurf npx -y @gitdealflow/mcp-signal ```

Refresh schedule. The pipeline runs every Monday morning. CDN caches at the edge for 24 hours; agents that hit the API mid-week get the most recent Monday data. The /api/changelog.json endpoint reports schema and pipeline changes.

If you want to compute it yourself, the methodology page documents the GitHub REST API endpoints (/repos/{org}/{repo}/stats/commit_activity, /contributors), the filtering rules, and the four-category signal-classification thresholds. The SSRN preprint at id 6606558 formalizes the approach.

The JSON response shape. The bulk endpoint returns a flat array of startup objects rather than a nested envelope, so it loads directly into a dataframe or a script without an unwrap step. Each record carries commitVelocity, commitVelocityChange, contributors, signalType, stage, sector, and recentRepos, and the output is sorted so the fastest-accelerating orgs come first. A single curl is therefore enough to generate a ranked shortlist.

No auth, no key. Every surface is free and unauthenticated, so there is no token to provision and no dashboard to log into. That makes the API a good fit for scripts and scheduled jobs that should not hold secrets. The only practical constraint is polite per-IP rate protection, so a well-behaved weekly pull never needs to think about quotas.

The MCP server. The @gitdealflow/mcp-signal package on npm exposes six read-only tools: get_trending_startups, search_startups_by_sector, get_startup_signal, get_signals_summary, get_scout_receipts, and get_methodology. Because they are read-only and require no credentials, they can be added to Claude Desktop, Claude Code, Cursor, or Windsurf and queried conversationally with nothing beyond the install step.

Other agent surfaces. The same panel is reachable through an MCP-over-HTTP transport, an A2A endpoint, an NLWeb endpoint, and a function-calling API, plus an RSS feed for the digest. The OpenAPI 3.1 spec at /api/openapi.json describes every callable route, so code generators can produce typed clients for the language you already use.

Reading the numbers correctly. The field that matters most is commitVelocityChange, the percentage delta against the prior 14-day window, normalized against each org's own baseline so a seed team and a Series B team rank on the same scale. The methodology behind it is validated against 219 startup-period observations, with a preprint on SSRN under abstract id 6606558, and the signal has historically surfaced breakout teams 3-6 weeks before fundraise announcements, a data range of 21 to 47 days with a median near 31 days.

Where the data comes from. The panel is derived from public GitHub activity using the standard REST endpoints documented on the methodology page, /repos/{org}/{repo}/stats/commit_activity and /contributors, combined with a filtering step over the 350+ startups in 15 sectors. The pipeline runs every Monday, and CDN caching holds each result for 24 hours, so a mid-week caller sees the most recent Monday snapshot. Several quarterly periods of history are stored, and per-period pages follow the pattern /startups-to-watch/{sector}-{period}, while the 14-day window slides forward seven days each weekly run. Higher granularity, daily data, is not exposed in the current panel; for arbitrary-repo tracking the GitHub REST API directly is the right tool.