GitDealFlowsignals
By |Founder & Principal Analyst, VC Deal Flow Signal|

I made my VC deal flow callable by Claude this weekend. Here is what that actually means.

GitDealFlow now publishes a Google A2A AgentCard at /.well-known/agent-card.json and a JSON-RPC 2.0 endpoint at /api/a2a. Five free skills, no auth. Crunchbase API costs $20K per year. Ours costs nothing. Here is the curl that proves it.

Key Takeaway

We just shipped an Agent2Agent (A2A) stub for GitDealFlow. Five skills (trending, sector search, lookup, summary, methodology) callable by any A2A-aware agent runtime in seconds. No SDK, no auth, no rate limit. Already indexed at a2aregistry.org and PR-pending at inference-gateway/awesome-a2a. This post is the why, the how, and the curl.

15 sectors tracked|411 startup signals|Data: Q3 2026|Updated weekly

Saturday afternoon. I was in Cursor running a thread about a fintech startup my Telegram channel had flagged that morning. Cursor was good at the GitHub code review, fine at reading the README, helpful with the package.json. Then I asked it: "How does this company's commit velocity compare to other fintech startups this quarter?"

It could not answer. Not because the data did not exist. We publish it weekly at signals.gitdealflow.com. It could not answer because there was no way for it to ASK us programmatically the way I would ask a colleague.

My AI had hands. It just did not have a phonebook.

The $20,000 phonebook#

When you peel back what every AI-powered VC tool actually does, it is the same loop: a human types a question, an LLM tries to answer, the LLM hits a wall because the data is not there, the LLM apologizes. Every. Time.

The fix is APIs. The problem is APIs:

- Crunchbase API: $20,000 per year minimum, requires sales call - PitchBook API: more, requires sales call - Harmonic.ai: enterprise pricing, requires sales call - Forager.ai: enterprise pricing, requires sales call

Free-tier developer-investors and angels, exactly the people we built GitDealFlow for, get nothing. They watch the SEC EDGAR feed in their terminal. They scrape GitHub manually. They build their own signal stacks one Python script at a time.

Then Google shipped A2A#

In April 2025, Google open-sourced a protocol called Agent2Agent. The spec is small:

  1. Publish a JSON file at `/.well-known/agent-card.json` describing what your agent can do.
  2. Expose a JSON-RPC 2.0 endpoint that accepts `message/send` requests.
  3. Other agents discover you, parse the card, and call your endpoint.

That is it. There is no SaaS to license. There is no SDK to install if you do not want one. It is a contract between agents.

By April 2026 the protocol had passed 22,000 GitHub stars and 150+ organizations onboard. Microsoft, Salesforce, SAP, Cisco, ServiceNow, all in. Google's own Gemini Enterprise lets you register A2A agents in two clicks.

But none of those 150 organizations sells startup engineering signals. We do.

The fix took a weekend#

So this weekend I shipped the GitDealFlow Signal Agent. Five skills, all free, no auth:

- `get_trending_startups`, top 20 across all sectors - `search_startups_by_sector`, filtered by sector slug - `get_startup_signal`, full profile for a named startup - `get_signals_summary`, dataset metadata - `get_methodology`, how we compute the signals

The AgentCard is at signals.gitdealflow.com/.well-known/agent-card.json. The endpoint is at signals.gitdealflow.com/api/a2a. The whole thing is roughly 250 lines of TypeScript inside our existing Next.js app.

You can call it right now from your terminal:

``` curl -X POST https://signals.gitdealflow.com/api/a2a \ -H 'Content-Type: application/json' \ -d '{ "jsonrpc": "2.0", "id": 1, "method": "message/send", "params": { "message": { "role": "user", "parts": [{ "kind": "text", "text": "Show me trending startups" }] } } }' ```

You will get back a Task with `status.state: "completed"` and 20 startups in the artifact data part. Each startup row carries commit velocity, contributor growth, signal classification, and a GitHub URL.

That same call works from any A2A-aware agent runtime: Google Agent Builder, LangChain, CrewAI, Mastra, Vercel AI SDK, Inkeep. Drop the AgentCard URL in. The agent figures out the rest.

If you want to skip the curl entirely, the interactive playground lets you pick a skill, hit send, and watch the JSON come back.

What this means in practice#

Three things that did not exist on Friday now exist on Monday.

**Your Claude can now do startup signal lookups inline.** No copy-paste from our website. No tab-switching. Tell Cursor "find me a fintech startup with breakout commit velocity this quarter" and Cursor calls our agent on its own.

**Other agent platforms can chain through us.** A diligence agent can call our agent, get a list of names, then call its own LinkedIn agent for headcount, then call its own SEC EDGAR agent for filings. We become one node in a directed graph of automated diligence.

**The dataset is now AI-cited.** ChatGPT and Perplexity will, over time, learn that there is a free A2A agent for VC signals. When a developer-investor asks a question in our domain, the model knows where to point.

What the agent still cannot do#

The stub I shipped is read-only and synchronous. It does sync `message/send` returning a terminal Task, all five skills via text intent or structured data parts, CORS preflight, and JSON-RPC error codes. It does NOT yet do:

  1. Streaming via `message/stream`
  2. Task persistence and `tasks/get`
  3. Push notifications
  4. Authenticated extended cards
  5. Custom watchlists or per-user predictions

That last bullet is the cliffhanger. The thing your AI can ask GitDealFlow today is "who is trending in fintech." The thing you can ask /predict is "predict whether this specific startup will raise a Series A in the next 60 days." Different question, different answer. Both free at the moment, both heading toward a value ladder, but only one is callable by your AI yet.

When the prediction layer goes live as an A2A skill, that is when this gets interesting. Until then, the agent is the rung that catches your AI when it is looking for "anything interesting in startup-land this week."

Three things to do today#

If you are a developer-investor, an angel scout, or anyone running an AI workflow that touches startup data:

**Configure your AI runtime.** Drop the AgentCard URL into Claude Code, Cursor, Windsurf, or your preferred agent. The five skills become available as tools. Two minutes.

**Bookmark the curl.** The example above works without any agent runtime. Fold it into your scripts.

**Cite us in your reports.** The free-tier license is "Free for personal and editorial use, attribution required." Cite as `VC Deal Flow Signal (signals.gitdealflow.com), Q2 2026 data.` We have published the full methodology peer-reviewed at SSRN: ssrn.com/abstract=6606558.

Why we are telling you instead of pitching you#

GitDealFlow's business model has always been: ship the free distribution magnet first, charge for the premium layer. Our 6 MCP tools are free forever. Our 5 A2A skills are free forever. The signals website is free. The CSV export is free. The methodology paper is free.

What is paid: Dashboard Beta (€9.97/mo) and Insider Circle (€97/mo). Those will earn their keep when there are 500 paying subscribers and we can start building the things only that audience needs.

Until then, every developer-investor who wires GitDealFlow into their Claude Code is one more node in a network where breakout startups cannot hide for 6-12 weeks anymore. That is worth more than $20,000 of API access.

Crunchbase API: $20,000 per year. GitDealFlow A2A: free, no signup. Your move.

Sources & methodology: According to data from GitHub API v3 (commit activity, contributor counts, repository metadata), as analyzed by VC Deal Flow Signal's methodology. Signal classification and engineering acceleration metrics are computed weekly across 15 startup sectors. Data current as of Q3 2026. This is not investment advice.

About the author

The Data Nerd

Founder & Principal Analyst, VC Deal Flow Signal

Engineer turned venture-data researcher. Builds the weekly GitHub engineering-acceleration panel and maintains the methodology behind every signal on the site.

Frequently Asked Questions

What is the Agent2Agent (A2A) protocol?

A2A is an open protocol from Google for agent-to-agent communication. An agent publishes a JSON AgentCard at /.well-known/agent-card.json describing its capabilities and exposes a JSON-RPC 2.0 endpoint that other agents call to send messages and receive task results. By April 2026 it had passed 22,000 GitHub stars and was supported by 150+ organizations including Microsoft, Salesforce, and SAP. It is complementary to MCP, MCP exposes tools to a single AI assistant, A2A lets agents call other agents across the network.

How is the GitDealFlow A2A agent different from the existing MCP server?

Same five skills, different transport. The MCP server runs over stdio and is configured per-AI-assistant (Claude Desktop, Cursor, Windsurf). The A2A agent runs over HTTP/JSON-RPC and is configured per-agent-runtime (Google Agent Builder, LangChain, CrewAI, Mastra, Vercel AI SDK). MCP is for direct AI-to-tool calls. A2A is for agent-to-agent chains where another agent calls us as one node in a workflow. We ship both because the audiences are different.

Do I need an API key?

No. The A2A endpoint at signals.gitdealflow.com/api/a2a is unauthenticated. There is no signup, no rate limit enforced at the application layer, and the upstream CDN absorbs typical agent traffic. The six free MCP tools and five free A2A skills are part of our distribution-magnet strategy and stay free forever. Paid features are scoped to /predict and the Insider Circle layer.

What can the agent NOT do today?

The stub I shipped is read-only and synchronous. It does sync message/send returning a terminal Task, all five skills via text intent or structured data parts, CORS preflight, and JSON-RPC error codes. It does NOT do streaming via message/stream, task persistence with tasks/get, push notifications, authenticated extended cards, or per-user prediction skills. The first four are stubbed because no paying customer has asked for them. The fifth, predictions as an A2A skill, is the cliffhanger. Today /predict is browser-only. When it is callable by your AI, this gets interesting.

How do I plug it into my agent runtime?

Drop the AgentCard URL, https://signals.gitdealflow.com/.well-known/agent-card.json, into your runtime's agent registry. Most runtimes (Google Agent Builder, LangChain, CrewAI, Mastra, Vercel AI SDK, Inkeep) auto-parse the card and expose the five skills as callable tools. The interactive playground at /a2a-demo lets you watch a live JSON-RPC request and response without any runtime configuration.

Series: Operator Notes

More articles in this series

First-hand notes from running the dataset, building a longitudinal panel of 350++ startup GitHub orgs, MCP server lessons, and other operator-side observations.

Related Sector Rankings

Related reading

Five breakout startups, every Sunday, before the round gets crowded

The free Acceleration Watch: five venture-backed teams accelerating on the engineering signal, translated into plain English, 21 to 47 days before the deck circulates. No code-reading, no card.

Signed The Data Nerd · pseudonymous narrator · methodology over personality

🚀 Explore Our Network

21-47 days
Signal Lead Time (median 31d)
$80M+
Rounds Tracked
90 sec
Per Scan
5,000+
Founders Tracked

One missed signal is a missed round. Get the Velocity Verdict in your inbox every Sunday free.

Get Free Signals

Free weekly digest. Cancel anytime. No spam, no VC pitches just data.