For AI Agents
Let Your Agent Do the Scouting.
Connect Claude, Cursor or your own agent to AlphaScout and it can pull the startups that fit your thesis, read the evidence behind each one and compare them, without scraping a single page. Every answer comes back with its sources.
Five Layers, and Where We Stand
An agent-ready product isn't one feature. It's five layers, each built on the one below: agents can read it, find it, call it, talk to it, and trust it. Here is AlphaScout on each, including what we haven't built.
- 5TrustworthyKeys you control, read-only tools, hard limits, and third-party text kept from steering your agent.In Place, With Limits
- 4ConversationalAn agent of our own that other agents delegate tasks to (A2A).Not Offered, on Purpose
- 3CallableAn MCP server with seven task-shaped tools, and a REST API with an OpenAPI file.Live
- 2DiscoverableA server card at the well-known paths, llms.txt, a sitemap and a clear robots policy.Live
- 1ReadableServer-rendered pages and schema.org data an agent can read without running a script.Live
1 · Readable
Every public page is rendered on the server, so an agent that doesn't run JavaScript still gets the whole page: the picks, the scoring rules, pricing, the docs and the blog. The home page carries schema.org data for the organization, the website and the product with its price, and every blog post is marked up as an article. URLs are stable, so an agent that cites a page today finds it tomorrow.
2 · Discoverable
An agent arriving at tryalphascout.com learns what it may read and what it can call in one or two requests:
| Path | What it tells an agent |
|---|---|
| /robots.txt | Crawlers and agents acting for a person may read the public site. The product and its data are never crawled; they're reached through the MCP server or the API with a key. |
| /.well-known/mcp.json | The MCP server card: endpoint, transport, protocol versions, how to authenticate, limits and every tool with its annotations. The same card is at /.well-known/mcp/server-card.json, since the discovery path isn't settled yet. |
| /api/v1/openapi.json | The REST API as OpenAPI, for frameworks that build tools from it. |
| /llms.txt | A plain index of the pages that matter, with what AlphaScout is and costs. |
| /sitemap.xml | Every public page, with honest last-modified dates. |
3 · Callable
This is where the value is. The MCP server at https://tryalphascout.com/api/v1/mcp speaks the streamable HTTP transport, statelessly: one request, one answer, no sessions to hijack. Its tools are shaped around the questions an investor asks, not around our database tables, so an agent needs few calls to get to an answer:
| Tool | What it does |
|---|---|
| list_hidden_gems | Ranked hidden gems: new launches nobody has backed yet, best first. |
| search_companies | Every tracked early company, filtered and sorted. |
| get_company | One company's full profile with its evidence and funding. |
| similar_companies | Comparable companies, with the reasons each one matches. |
| latest_releases | What companies just shipped, one entry per company. |
| latest_signals | Raw evidence, newest first. |
| scoring_rules | The scoring rules exactly as the engine applies them. |
Every tool is marked read-only, non-destructive and idempotent, so a client never needs to stop and ask before calling one. Results are sized for a model's context, and a bad argument comes back as an error that says how to fix the call. Add the server to your client with an API key:
claude mcp add --transport http alphascout https://tryalphascout.com/api/v1/mcp \
--header "Authorization: Bearer as_live_…"Then ask in plain language: "Which developer-tools startups in Europe launched this week, and what's the evidence?" The full reference is in the MCP docs, and the same data is on the REST API for agents that don't speak MCP.
4 · Conversational
We don't run an Agent2Agent (A2A) endpoint, and that's a choice. A2A is for handing a long task to another agent with its own judgement. What an agent wants from AlphaScout is data with evidence: which companies fit, why, and where it says so. That is what tools are for, and wrapping them in a second protocol would add ceremony, not capability. If we ever offer an agent that does real delegated work, it will sit on top of these same tools.
5 · Trustworthy
- Keys you control. Workspace owners create API keys in the app, can limit each to their servers' addresses and give it an expiry, and revoke it any time. Keys are stored only as hashes.
- Read-only by design. No tool can change, delete or send anything. The worst a misbehaving agent can do is read.
- Hard limits on our side. 60 requests a minute and 10,000 a day per key, one message per request, a 64 KB body cap. A key used from many addresses at once, or walking through every profile, is paused and its owner is told.
- Third-party text stays data. Company names and descriptions come from company websites, which could try to address your agent. Invisible characters are stripped from every result, and the server tells agents to treat that text as data, never as instructions.
- Companies, not people. No personal contact details are stored or returned. See Security for the rest.
The Checklist, Honestly
| Item | Status | Detail |
|---|---|---|
| Key content in server-rendered HTML | Done | Every public page |
| schema.org JSON-LD | Done | Organization, WebSite, SoftwareApplication with its price, and every blog post |
| robots.txt separates crawlers from agents acting for a person | Done | Both may read the public site; neither may crawl the product |
| Remote MCP server with task-shaped, honestly annotated tools | Done | Seven tools, all read-only |
| MCP server card at the proposed well-known paths | Done | /.well-known/mcp.json and /.well-known/mcp/server-card.json |
| OpenAPI document with descriptions | Done | /api/v1/openapi.json |
| llms.txt | Done | /llms.txt |
| Per-key rate limits, audit trail and abuse detection | Done | 60 a minute, 10,000 a day; unusual use pauses the key |
| Untrusted text labelled and cleaned in tool output | Done | Invisible characters removed; agents told to treat it as data |
| OAuth 2.1 sign-in with protected-resource metadata | Not Yet | Not yet: keys only, so chat connectors that need OAuth can't connect |
| Listed in the official MCP Registry | Not Yet | Not yet |
| A2A Agent Card | Not Yet | Not offered: we run tools, not a delegating agent |
| Agent payments (AP2, ACP, x402) | Not Yet | Not offered: plans are bought by people, keys are issued per workspace |
Questions about connecting your agent? Talk to us, or read the developer docs.
Free Sample
The Next Breakout Is Already in the Data.
Last week's hidden gems are free to browse, no account needed. When you want this week's, picked for your thesis, your list is waiting.