jintel-sdk

# Jintel SDK Public SDK monorepo for the [Jintel](https://api.jintel.ai) intelligence API — a GraphQL endpoint that unifies market data, news, regulatory filings, sanctions screening, macro indicators, and more behind one key. Get an API key at [api.jintel.ai](https://api.jintel.ai). ## Packages | Package | Install | Use | |---|---|---| | [`@yojinhq/jintel-client`](packages/client) | `npm i @yojinhq/jintel-client` | TypeScript SDK — typed queries, Zod schemas, response cache | | [`@yojinhq/jintel-cli`](packages/cli) | `npm i -g @yojinhq/jintel-cli` | `jintel` command for terminals, shell scripts, CI | | [`@yojinhq/jintel-mcp`](packages/mcp) | `npx -y @yojinhq/jintel-mcp` | MCP server for Claude Desktop, Claude Code, and other MCP clients | | [`jintel` plugin](packages/claude-plugin) | `claude plugin install jintel@YojinHQ` | One-shot Claude Code plugin: MCP server + skill + slash commands | All four wrap the same GraphQL API. Array sub-graphs accept a per-field `filter` argument — generic date/limit/sort via `ArrayFilterInput`, plus domain-specific inputs for SEC filings (`FilingsFilterInput`), risk signals (`RiskSignalFilterInput`), futures (`FuturesCurveFilterInput`), and options chains (`OptionsChainFilterInput`). See [`packages/client/README.md`](packages/client/README.md#filtering-array-sub-graphs) for examples. ## Auth modes Every package supports two auth modes — set exactly one: - **Bearer (API key)** — `JINTEL_API_KEY=jk_live_…`, get one at [api.jintel.ai](https://api.jintel.ai). Org-billed via plan + credit pool. - **x402 wallet (keyless)** — `JINTEL_WALLET_PRIVATE_KEY=0x…`, pays per query in USDC on Base (~$0.015 floor). Optional `JINTEL_X402_MAX_VALUE` caps spend per query (atomic USDC, default `1000000` = $1). Agent-native: no signup, no Bearer — the payer wallet auto-identifies the caller. See [`packages/client/README.md`](packages/client/README.md#agent--pay-per-query-x402-v2). `JINTEL_API_KEY` wins if both are present. ## Quick start ```bash # Either mode works: export JINTEL_API_KEY=jk_live_your_key_here # Bearer # export JINTEL_WALLET_PRIVATE_KEY=0xabc... # x402 wallet (USDC on Base) # CLI npm i -g @yojinhq/jintel-cli jintel quote AAPL MSFT # MCP (Claude Desktop ~/Library/Application Support/Claude/claude_desktop_config.json) # "mcpServers": { "jintel": { "command": "npx", "args": ["-y", "@yojinhq/jintel-mcp"], # "env": { "JINTEL_API_KEY": "..." } } } # Wallet mode: swap "JINTEL_API_KEY" for "JINTEL_WALLET_PRIVATE_KEY" in the env block. # Claude Code plugin claude plugin marketplace add YojinHQ/jintel-sdk claude plugin install jintel@YojinHQ ``` ## Local development ```bash pnpm install pnpm build # build all packages pnpm typecheck pnpm test ``` ## Publishing ```bash pnpm publish:client patch # or minor / major pnpm publish:cli patch pnpm publish:mcp patch ``` ## Benchmark [`benchmark/`](benchmark/README.md) is a measurement framework that compares LLM-with-tools configurations on a corpus of financial-research queries. Variants include `bare` (no tools), `anthropic-web-search`, `jintel-mcp` (the published `@yojinhq/jintel-mcp`), and `jintel-cli` (the published `@yojinhq/jintel-cli` over a `bash` tool). ```bash pnpm bench --model claude-haiku-4-5 --variant all pnpm bench:report ``` Latest published results: [`benchmark/results/`](benchmark/results/README.md). On the 15-question corpus, Haiku-with-`jintel-mcp` (66.7%) beats Opus-with-web-search (60%) at roughly half the latency, and Opus-with-`jintel-mcp` leads at 86.7%. See [`benchmark/README.md`](benchmark/README.md) for the full CLI, corpus schema, and grader rules. ## License MIT — see [LICENSE](LICENSE). ## Links - API docs: https://api.jintel.ai/docs - Homepage: https://api.jintel.ai - Main (private) repo: Jintel API server lives in a separate repo owned by Othentic Labs.