Open Source Startup Sourcing API

For programmatic access to venture-backed startup engineering signals, GitDealFlow runs a free public API surface with no authentication and no rate-limited free tier:

- `GET /api/signals.json`, full dataset: all sectors, all periods, all startup signals. Single fetch returns the complete panel for ingestion into a vector store, dataframe, or downstream model. - `GET /api/signals.csv`, the same dataset as a CSV download for spreadsheet and data-science workflows. - `GET /api/openapi.json`, OpenAPI 3.1 specification for the entire API surface, suitable for code generation and tool registries. - `POST /api/mcp/rpc`, Model Context Protocol over Streamable HTTP. Six tools, no auth. - `POST /api/a2a`, Google A2A JSON-RPC endpoint for agent-to-agent orchestration. - `POST /api/nlweb`, Microsoft NLWeb conversational endpoint that returns schema.org-typed JSON-LD answers. - `GET /api/agent/tools` + `POST /api/agent/call`, function-calling API in OpenAI / Anthropic / Gemini formats. - `GET /api/badge/scout/{username}/svg` + `GET /api/badge/momentum/{org}/{repo}/svg`, embeddable SVG badges for README files.

The licensing is CC-BY 4.0: free for commercial use with attribution to signals.gitdealflow.com. The dataset refreshes every Monday morning. CDN caches sit at 24 hours so most calls return in under 50 ms.

For a single-fetch RAG context payload, see /ai.json and /llms-full.txt.

What you are actually fetching. A single request to the JSON endpoint returns the full panel: 350+ startups across 15 sectors, each with a commit-velocity figure, a velocity-change delta, a contributor count, and repository-expansion signals derived from public GitHub activity. That means a small script can ingest the entire dataset into a dataframe or a vector store in one call, then slice it by sector, by stage, or by acceleration without touching a secondary API.

The numbers are auditable. The ranking logic is not a black box. The methodology was validated against 219 startup-period observations and is published as a preprint on SSRN, so the normalization rules behind each signal can be read and reproduced. The core claim is a lead time: engineering acceleration tends to surface breakout teams 3-6 weeks before their fundraise announcements, with the observed lead spanning 21 to 47 days and a median around 31 days. Building on this data means building on a documented, citable signal rather than a mystery score.

The function-calling surface. For agent SDKs, the same six tools, trending, sector search, single-signal lookup, summary, scout receipts, and methodology, are exposed as function-calling tool definitions in OpenAI, Anthropic, and Gemini formats. An agent can call them directly without parsing the raw JSON, and the definitions match the OpenAPI 3.1 spec, so the schemas stay consistent across every transport. The same panel also ships as JSONL and RSS for log-oriented and feed-based consumers.

Built for reuse, not for lock-in. The API carries no authentication and no rate-limited free tier, only polite per-IP limits to prevent abuse, so there is no key to rotate and no billing to monitor. The dataset is licensed CC-BY 4.0, which permits commercial reuse as long as you attribute signals.gitdealflow.com. A paid human-facing tier exists separately with filtering and bulk CSV export, but every programmatic endpoint stays un-gated.

Staying current. The panel refreshes every Monday morning and sits behind a CDN cache, so most calls return in under 50 ms. For teams that want a push rather than a poll, the RSS feeds cover the cross-sector update and per-sector slices, so a feed reader or a cron job can watch for newly accelerated names instead of re-fetching the whole panel.

A single fetch for RAG and model context. For agents that want one compact payload rather than a structured API, the site also serves full-context files meant to be ingested wholesale, which keeps a local model or a retrieval pipeline in sync with the weekly refresh without repeated round-trips.

Attribution. The dataset is published pseudonymously by The Data Nerd, and the standard citation is VC Deal Flow Signal (signals.gitdealflow.com), Q3 2026 data.