For AI agent builders · Pay-as-you-go · No subscription
Your AI agent shouldn’t re-read Crunchbase.
Plug into 350+ GitHub orgs of live signal.
If you’re shipping a deal-flow agent, an investor-briefing assistant, or a startup-research tool, the bottleneck isn’t the LLM, it’s the data layer. We sell the data layer. Six free MCP tools cover discovery and shallow lookup. The seventh, get_deep_signal runs at €0.19/call with misses free. Zero rate limits inside quota. ~200ms median.
Start with the highest-intent routes
Use this page when you want the builder-specific offer. But if your real question is proof, MCP workflow fit, or buyer-side evaluation, start with the sharper pages first.
The single belief
Most agent stacks ship with scraped Crunchbase rows as the startup-research backbone. Crunchbase tells you the day they announced. We tell you 47 days before.
That’s the only durable difference between an agent that briefs investors after the fact and one that briefs them before the round closes. Same LLM, same prompt, completely different output, because the input is forward-looking instead of lagging.
“The deck lags the code by 21 to 47 days.” The Data Nerd
Six free MCP tools, distribution magnet, never gated
Start here. Free, no signup, no API key.
The free six cover discovery and shallow lookup. Most agents never need to upgrade, the seventh tool is for when you’ve identified a target and need report-grade output for an investor brief.
get_trending_startups
Top N startups across all 15 sectors, ranked by 14-day commit-velocity acceleration.
search_startups_by_sector
Filter the 350+ org universe by sector slug + stage + geography. Returns ranked list.
get_startup_signal
Shallow signal for one named org, current acceleration percentile, signal type, last refresh.
get_signals_summary
Aggregate signal counts across the entire universe, hiring bursts, infra buildouts, shipping sprints, framework migrations.
get_scout_receipts
Backwards-looking taste grade for any GitHub username. Same engine that powers /receipts.
get_methodology
Open methodology spec, every signal definition, the SSRN abstract, the Zenodo dataset link, the regression code.
Drop-in MCP config (Claude / Cursor / Windsurf / any host)
{
"mcpServers": {
"gitdealflow": {
"command": "npx",
"args": ["-y", "@gitdealflow/mcp-signal"]
}
}
}No env var, no API key. Restart your agent host and the six tools appear in the catalog.
The seventh tool, pay-per-call deep enrichment
get_deep_signal · €0.19/call · misses free
Returns a composite 0-100 score, in-sector rank, plain-English thesis, top 3 comparables, multi-period history, and a citation string. Hits charge one credit. Misses (org not in our universe) return { found: false } with a suggestion and charge zero. Credits never expire.
HTTP, direct from any agent runtime
curl -X POST https://signals.gitdealflow.com/api/agent/deep-signal \
-H "Authorization: Bearer gdf_v2.cus_xxx.<your_hmac>" \
-H "Content-Type: application/json" \
-d '{"name":"supabase"}'Starter pack, 100 credits
€19
= 100 calls · €0.19/call
Stripe-hosted checkout. API key delivered in ~30s. Top up any time, no expiration.
Crypto-native alternative, x402 / USDC on Base
Fully autonomous agents with their own wallet can skip the credit pack and pay per call in USDC on Base via the x402 protocol. No signup, no API key, same payload, $0.19/call settled per request via HTTP 402. Misses still free.
POST https://signals.gitdealflow.com/api/agent/deep-signal/x402
{ "name": "supabase" }
# 402 Payment Required → sign EIP-3009 → retry with X-PAYMENT headerWhat people are building with this
- Investor-briefing agents. Drop a Slack command into a fund’s internal channel, return a one-page brief on any startup with composite score, comparables, and the citation block ready to paste into an IC memo.
- Deal-flow assistants in Cursor. An angel writes 5-15 cheques a year. Their Cursor agent surfaces the week’s ranked list, deep-signals their three favourite names, drafts cold outbound, and queues it for review.
- Sector-research notebooks. Researcher writes a thesis prompt, agent calls get_signals_summary + a handful of get_deep_signal lookups, and returns a structured report with chart data and a reproducibility appendix.
- Portfolio-overlap detectors. Existing fund tooling pings the API on every new prospect to flag overlaps with the public-signal universe and returns the acceleration delta vs. the closest comparable already in the fund’s book.
Practical answers in five lines
- •Methodology is open, SSRN abstract 6606558, Zenodo dataset with DOI, regression code on GitHub.
- •Output license is permissive, drop the data into briefs, products, or training pipelines. Attribution requested, CC BY 4.0 on the dataset, MIT-style on responses.
- •Zero rate limits inside your quota. Burst freely.
- •~200ms median latency on get_deep_signal. HMAC-keyed, fluid compute, no DB lookup on hot path.
- •Every paid tier (and the free MCP) supports the same toolset switching surfaces does not change tool names or signatures.
FAQ
Why a separate page for builders?
Because this page is not about reading the signal. It is about wiring the signal into an agent, workflow, or internal tool. Same data, different job.
Are the six free tools really free forever, or is this a trial?
Forever. The free MCP server is the public entry layer. Credits only apply to the seventh tool (get_deep_signal) and its HTTP twin (POST /api/agent/deep-signal). The free six stay free permanently.
What does get_deep_signal return that the free tools don't?
A composite 0-100 score weighted across velocity / contributor growth / repo novelty, in-sector rank and percentile, a one-sentence plain-English thesis you can drop into a memo, top 3 sector comparables with signal type, up to 6 prior periods of velocity history, and a citation string. The free get_startup_signal returns just the current percentile and signal type. The deep tool is what you call when you've identified a target and need report-grade output.
How does the credit metering work?
One credit per match. €19 buys 100 credits = €0.19/call. If you query an org we don't track ('found: false'), the call is free and your balance is unchanged. Credits never expire, buy when you need to, use when you need to. Top up any time. No subscription, no monthly minimum, no auto-renewal. The HMAC API key stays the same key forever; topping up just adds credits to its balance.
What's the latency target?
~200ms median for get_deep_signal. The endpoint is HMAC-keyed (no DB lookup on the hot path), runs on Vercel Functions in fluid-compute mode, and pre-computes scores nightly. Inside your quota there are no rate limits, burst freely.
Will this stay €0.19/call?
Yes for now. We may add bulk-pack pricing (1k+ packs at lower per-call rates) once volume justifies it. Founding builders who buy at €0.19/call will keep that rate on top-ups indefinitely, same lock-in policy as the founding-member subscription tiers.
Can I see code examples?
Yes, three runtimes covered on /agents/credits: MCP config (Claude Desktop, Cursor, Windsurf, any host), HTTP curl, and balance check. The MCP server is the same npm package as the free version (@gitdealflow/mcp-signal); set GITDEALFLOW_API_KEY in env and the deep tool wakes up. The other six tools ignore the env var entirely.
Plug it in. Burn 100 credits in a weekend. Decide.
The free tools take three minutes to wire into Claude Desktop or Cursor. The paid tool takes one Stripe checkout and a bearer-token header. If your agent doesn’t produce sharper output after burning the first 100 credits, the credits never expire, and the free six tools stay live regardless.