Docs

Quick setup.

Remote MCP works without an account at https://buyapi.ai/api/mcp. Use the CLI when you want guided setup, local fallback, or private stack sync.

§ Setup

~/.claude/mcp_servers.json

{
  "mcpServers": {
    "buyapi": {
      "url": "https://buyapi.ai/api/mcp"
    }
  }
}

Once connected, ask your agent recommend an email api for a side project. The tool call happens automatically — no wrapper library, no CLI.

Local fallback for clients without remote MCP

{
  "mcpServers": {
    "buyapi": {
      "command": "npx",
      "args": ["-y", "buyapi", "mcp"]
    }
  }
}

§ Stack sync

The scanner inspects project files on your machine. Plain scan shows a local preview first and can ask before saving; dry-run never uploads. Sync requires an API key and saves detected tools to your private dashboard.

$npx buyapi setup cursor# or codex / claude-code

$npx buyapi scan# run from project dir

$npx buyapi login# for sync / higher limits

$npx buyapi scan --sync --yes# in project dir

Today the scanner uses rules bundled in the open-source package: package names, framework config files, common service directories, and safe `.env.example` key names. Source files and environment values are not uploaded; only tool IDs, categories, confidence, and detection notes are saved.

§ Tools

Seven MCP tools. Agents can discover vendors, fetch sourced detail, inspect evidence, find similar stacks, compare head-to-head options, run directional cost math, or jump straight to a complete stack recommendation.

vendors.resolveFirst stop for category-specific vendor recommendations. Take a user question and optional category; return ranked vendor IDs, or an explicit not-in-corpus result.

Parameters

query: The user's question or task; required
category: database · auth · hosting · payments · email · ui-components; optional

Returns

A short list of vendor IDs plus, for covered category queries, a top-3 decision matrix with fit labels, confidence, tradeoffs, cost notes, freshness, and sources.

vendors.detailsFollow-up for one known vendor — pricing tiers, limits, feature grid, gotchas, comparisons, and source provenance.

Parameters

vendorId: BuyAPI vendor ID (e.g., /database/neon); required
query: The specific question to tune the response; optional

Returns

A structured vendor profile with pricing, limits, features, known issues, comparisons, company facts, claims, freshness, and confidence.

vendors.evidenceTrust follow-up: fetch reviewed evidence rows for a vendor, category, stack, or comparison.

Parameters

subjectType: vendor · category · stack · comparison; required
subjectId: e.g. /database/supabase or database; required
limit: Maximum evidence rows; optional

Returns

Reviewed evidence rows with source type, summary, stance, confidence, source URL, author, and observed date.

stacks.findSimilarFind reviewed public stack profiles related to a vendor, or recent curated stack examples.

Parameters

vendorId: Optional BuyAPI vendor ID; optional
limit: Maximum stack profiles; optional

Returns

Public stack profiles with project summary, audience, stage, tools, roles, and confidence.

vendors.compareCompare two or more already-known vendors for a stated workload or decision.

Parameters

vendorIds: BuyAPI vendor IDs, e.g. /database/convex; required
query: The user's decision context; required
workload: Explicit workload assumptions; optional

Returns

A decision matrix with fit, tradeoffs, estimated monthly cost, confidence, freshness, and sources.

vendors.estimateCostRun directional BuyAPI cost estimates from reviewed pricing claims and explicit workload inputs.

Parameters

workload: users, MAU, storage GB, email sends, auth MAU, regions, seats, payments volume, notes; required
vendorIds: Specific BuyAPI vendor IDs; optional
category: Estimate across a category in the current corpus; optional

Returns

Cost estimates with basis, assumptions, unknowns, confidence, and source provenance. Missing workload fields become assumptions, not fabricated precision. Verify exact billing in first-party docs, vendor CLIs, or vendor MCPs before purchase or production decisions.

stacks.recommendOne call, full stack. Give BuyAPI a project description, constraints, and optional workload; get one pick per layer plus structured reasoning.

Parameters

projectDescription: What you're building (a sentence or two); required
constraints: Budget, scale, compliance, existing tools; optional
workload: Explicit workload assumptions; optional
stackContext: Existing tools from a repo scan or saved stack; optional

Returns

The same core recommendation shape as /api/recommend: stack, decisionMatrix, costEstimate, assumptions, unknowns, alternatives, sources, and generatedAt.

§ Vendor IDs and workload hints

BuyAPI vendor IDs use /category/slug. Use exact IDs when you already know the tools you want compared.

/database/convex
/database/supabase
/database/neon
/auth/clerk
/hosting/vercel
/payments/stripe
/email/resend
/ui-components/shadcn-ui

The strongest prompts include real workload numbers: monthly active users, team seats, storage GB, email sends, auth MAU, regions, transaction volume, and budget ceiling. Use BuyAPI for the vendor decision, then pair it with Context7 or first-party docs for exact implementation APIs and version-specific code examples.

§ REST API

For non-MCP clients (scripts, web apps, your own agent frame). The same data; plain HTTPS.

GET/api/vendors/search

?query=...&category=...

→ { results, decisionMatrix? } where category queries can include the same top-3 fit matrix used by vendors.resolve

GET/api/vendors/[...slug]

e.g., /api/vendors/database/neon

→ Full VendorProfile object (see Tools → vendors.details for shape)

POST/api/recommend

body: { projectDescription, constraints?, workload?, stackContext? }

→ { stack, decisionMatrix, costEstimate, assumptions, unknowns, alternatives, sources, generatedAt }

GET/api/evidence

?subjectType=vendor&subjectId=/database/neon&limit=8

→ { evidence } reviewed evidence rows with source URLs, stance, confidence, and observed dates

GET/api/stacks/similar

?vendorId=/database/convex&limit=6

→ { stacks } public stack profiles with tools, roles, stage, audience, and confidence

All endpoints return JSON. No auth required at default rate limits — see below.

§ Rate limits

Anonymous

10 / min

Default. No setup, no account.

With key

60 / min

10/min is plenty for a single developer — an agent makes 1–3 calls per real question. If you're rate-limited, use a keyed MCP or CLI setup and retry after the response hint.