Strale — Trust And Quality Infrastructure For Ai Agents

Strale

Pay-per-call tools for AI agents — one API key or an x402 wallet. MCP server, REST API, TypeScript and Python SDKs.

What Strale is

Strale is a capability marketplace for AI agents. Agents call strale.do() at runtime to reach tools — company registry lookups, compliance checks, financial validation, Web3 security, and more — plus bundled solutions for multi-step workflows like a full KYB check or company due diligence, instead of the agent's own code hardcoding integrations and managing credentials for every data source it might need.

Each tool page lists inputs, output fields, price, limitations and source. Testing modes vary by tool. Read the current catalogue, free-tier list and country coverage from platform facts.

How to use it

MCP server (recommended: no install)

{
  "mcpServers": {
    "strale": {
      "type": "streamableHttp",
      "url": "https://api.strale.io/mcp?src=gh-readme",
      "headers": {
        "Authorization": "Bearer sk_live_your_key_here"
      }
    }
  }
}

Works with Claude Desktop, Claude Code, Cursor, and any MCP client supporting Streamable HTTP. For local stdio instead, or the full tool surface, see packages/mcp-server/README.md.

TypeScript SDK

npm install straleio

import Strale from "straleio";

const strale = new Strale({ apiKey: process.env.STRALE_API_KEY });
const result = await strale.do("eu-vat-validate", { vat_number: "SE556000000001" });

Full reference: packages/sdk-typescript/README.md.

Python SDK

pip install straleio

from straleio import Strale

strale = Strale(api_key="your_api_key")
result = strale.do("eu-vat-validate", {"vat_number": "SE556000000001"})

Full reference: packages/sdk-python/README.md.

Get started without an API key

A subset of capabilities (email/IBAN/crypto-address validation, DNS lookup, JSON repair, URL-to-markdown, and more) work with no signup, no API key, and no wallet — only an IP-based daily rate limit. Read the current list from GET /v1/platform/facts (free_tier_slugs), not from a number in this file. Get a key and trial credits at strale.dev when you need the rest of the catalog.

Framework integrations

Package	Registry	Description
strale-mcp	npm	MCP server — the full capability catalog via Claude, Cursor, any MCP host
straleio	npm	TypeScript/JavaScript SDK
straleio	PyPI	Python SDK
langchain-strale	PyPI	LangChain toolkit (StraleToolkit)
crewai-strale	PyPI	CrewAI integration — drop-in BaseTools for agents
strale-semantic-kernel	npm	Semantic Kernel plugin for .NET and TypeScript agents
composio-strale	PyPI	Composio integration — custom actions

Each package's own README is the reference for that integration; this file only points at them. Every distribution PR against an external framework repo, and every publish of one of these packages, follows CLAUDE.md's Distribution PR Integrity Protocol — a package claiming framework-native integration is verified against the published artefact, not the source tree, before the claim ships.

Web3

Web3 capabilities (wallet risk scoring, token honeypot detection, ENS resolution, DeFi protocol data, gas oracle, EU MiCA VASP verification, market sentiment) and bundled solutions for on-chain agents are available via the x402 payment protocol — pay per call with USDC on Base mainnet, no signup required. Not every capability is x402-eligible; read the live, dynamic subset from:

GET https://api.strale.io/x402/catalog?src=gh-readme
GET https://api.strale.io/.well-known/x402.json

Testing and source context

Testing modes vary by tool. Inspect current availability, test mode and last-tested time in public test signals, and read limitations and sources in the catalogue before use.

GET https://api.strale.io/v1/capabilities/vat-validate

Every capability publishes its limitations honestly, including coverage gaps. Methodology: strale.dev/quality.

IDE rules for safe data handling

Drop these into your project to give your coding agent security guidance for trust-sensitive data (IBANs, company registries, sanctions, PII):

• Cursor: docs/ide-rules/strale-compliance.mdc → .cursor/rules/
• Windsurf: docs/ide-rules/strale-compliance.windsurfrules → project root, or .windsurf/rules/ as .md
• Claude Code: the snippet at docs/claude-md-snippet.md → your project's CLAUDE.md

These prevent the most common vibe-coding vulnerabilities: client-side IBAN validation, hardcoded API keys, missing provenance metadata, and direct registry scraping.

Agent Skills & code examples

• Agent Skills — SKILL.md files for Claude Code, Cursor, Copilot, and Codex. Copy skills/strale/ into .claude/skills/, .github/skills/, or .cursor/skills/.
• Verify a company with Strale (Python)
• Validate IBANs — free, no API key (Python)
• Build an agent with Strale (TypeScript)
• Connect Strale to Claude (setup guide)

More

• strale.dev — homepage and sign-up
• strale.dev/docs — API reference
• strale.dev/pricing — pricing
• scan.strale.io — Beacon, a free agent-readiness scanner
• api.strale.io/mcp — MCP endpoint (Streamable HTTP)
• Examples — copy-paste examples for every integration

How the code is organised

strale/
├── apps/api/          Hono API server: routes, capability executors,
│                       Drizzle schema + queries, Stripe/matching/auth
│                       helpers. drizzle/ holds the migration history.
├── packages/           Published SDKs and framework integrations (see the
│                       table above) plus internal workspace packages.
├── manifests/          One YAML file per live capability: pricing, data
│                       source, field-reliability declarations, test
│                       fixtures. The onboarding pipeline
│                       (apps/api/scripts/onboard.ts) is the only
│                       sanctioned way a capability enters the system —
│                       see CLAUDE.md's Capability Onboarding Protocol.
├── design/             Design tokens as data (colors, type, spacing) —
│                       one active file per surface, candidates carry a
│                       status, promotion is a decision plus a file swap.
├── config/              Cross-cutting configuration as data, e.g. the
│                       environment-variable manifest every process.env
│                       read is registered against.
└── scripts/            Repo-root tooling: the checkers behind every
                        `npm run *:check` script, the session-handoff
                        gate, and one-off operator scripts.

apps/api/scripts/ and apps/api/src/capabilities/ carry the same idea at capability-authoring scope — read CLAUDE.md's "Adding New Capabilities" section before writing a new executor.

How the company runs the repository

Where truth lives. Product, state, and roadmap truth is migrating from Notion into this repository (docs/project/), but the migration is not complete: docs/project/START-HERE.md and the rest of docs/project/ are explicitly candidates (authority_active: false) until a founder- confirmed cutover. Until then, Notion (Project Home, the To-do & Build Plan, the Decisions DB) is authoritative, and CLAUDE.md / AGENTS.md say so at the top of every session. docs/README.md indexes every docs/ subtree with its current authority status; docs/project/STRUCTURE.md records exactly where the repo's layout still deviates from the migration's target and why.

Where work in flight lives. Multi-batch work is tracked in docs/programs/ (start at docs/programs/README.md): each program has a PROGRAM.md with a "Resume here" section and a machine-checked tracks.yaml. A session resuming a program reads only those two files and follows their pointers — chat history is never required.

How sessions work. CLAUDE.md is the canonical operating manual for Claude Code sessions; AGENTS.md is its condensed derivative for Codex-CLI sessions and points back at CLAUDE.md for anything that can drift rather than restating it. Every session ends through a handoff gate (scripts/handoff/handoff-check.mjs, installed as a git hook) that refuses to let a session leave uncommitted work, an unpushed branch, a stale worktree, or code changes with no resume surface for the next session. Batch work happens in an isolated git worktree, never the shared primary checkout — see WORKTREES.md.

What CI checks, by category, not by exhaustive list (see .github/workflows/ci.yml for the exact, current set): an ephemeral- Postgres integration lane for money- and audit-chain-critical behavior; typecheck and lint across apps/api and every published package; capability manifest structural and fixture-consistency gates; framework-package integrity (a published *-strale package must contain real code from the framework it claims to integrate with); a family of content-as-data registers, each with its own checker and generated index — research (docs/research/), design tokens (design/), the environment-variable manifest (config/), the model-id registry, the public claims register (docs/company/claims.yaml), program track registers (docs/programs/*/tracks.yaml), and this repository's own docs/ structure (docs:check, archive:index -- --check); and the session- handoff gate itself. Every one of these checkers plants a failure case in a throwaway fixture and proves it fails before the fix — see docs/company/LESSONS.md family F5.

Where history lives

archive/ holds closed, historical, and superseded material — see archive/README.md for what each subtree holds, generated (with hand- written prose above it) by npm run archive:index. handoff/ holds one file per session's end-of-session record; handoff/README.md is a generated, reverse-chronological index of all of them by date and stated intent. Neither directory is required reading to resume work — the program register and docs/project/START-HERE.md are.

License

MIT