Jintel
Model Context Protocol server for the Jintel intelligence API
Ask AI about Jintel
Powered by Claude Β· Grounded in docs
I know everything about Jintel. Ask me about installation, configuration, usage, or troubleshooting.
0/500
Reviews
Documentation
Jintel SDK
Public SDK monorepo for the Jintel intelligence API β 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.
Packages
| Package | Install | Use |
|---|---|---|
@yojinhq/jintel-client | npm i @yojinhq/jintel-client | TypeScript SDK β typed queries, Zod schemas, response cache |
@yojinhq/jintel-cli | npm i -g @yojinhq/jintel-cli | jintel command for terminals, shell scripts, CI |
@yojinhq/jintel-mcp | npx -y @yojinhq/jintel-mcp | MCP server for Claude Desktop, Claude Code, and other MCP clients |
jintel 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 β 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 for examples.
Auth modes
Every package supports two auth modes β set exactly one:
- Bearer (API key) β
JINTEL_API_KEY=jk_live_β¦, get one at api.jintel.ai. Org-billed via plan + credit pool. - x402 wallet (keyless) β
JINTEL_WALLET_PRIVATE_KEY=0xβ¦, pays per query in USDC on Base (~$0.015 floor). OptionalJINTEL_X402_MAX_VALUEcaps spend per query (atomic USDC, default1000000= $1). Agent-native: no signup, no Bearer β the payer wallet auto-identifies the caller. Seepackages/client/README.md.
JINTEL_API_KEY wins if both are present.
Quick start
# 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
pnpm install
pnpm build # build all packages
pnpm typecheck
pnpm test
Publishing
pnpm publish:client patch # or minor / major
pnpm publish:cli patch
pnpm publish:mcp patch
Benchmark
benchmark/ 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).
pnpm bench --model claude-haiku-4-5 --variant all
pnpm bench:report
Latest published results: benchmark/results/.
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 for the full CLI, corpus
schema, and grader rules.
License
MIT β see 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.
