agent-toolbelt

agent-toolbelt

Official SDK for Agent Toolbelt — a suite of focused API tools for AI agents and developers.

Typed client + LangChain tool wrappers for stock research (investment thesis, earnings analysis, insider signals, valuation, bear/bull, head-to-head comparison, moat analysis, watchlist scan), schema generation, text extraction, token counting, CSV conversion, Markdown conversion, URL metadata, regex building, cron expressions, address normalization, color palette generation, brand kit creation, meeting action item extraction, prompt optimization, web summarization, and more — 28 tools total (8 stock + 20 utility).

Install

npm install agent-toolbelt

Get an API key

curl -X POST https://www.agenttoolbelt.live/api/clients/register \
  -H "Content-Type: application/json" \
  -d '{"email": "[e-mail]"}'

Typed Client

import { AgentToolbelt } from "agent-toolbelt";

const client = new AgentToolbelt({ apiKey: process.env.AGENT_TOOLBELT_KEY! });

Stock Research Tools

Seven tools that pull live data from Polygon.io, Finnhub, and Financial Modeling Prep, then synthesize Motley Fool-style analysis. Each call returns structured JSON with verdict, key drivers, and what to watch. US-listed equities only (NYSE, NASDAQ, AMEX).

Investment Thesis

const result = await client.stockThesis({
  ticker: "NVDA",
  timeHorizon: "3-5 years",
});
// result.verdict        → "bullish" | "neutral" | "bearish"
// result.oneLiner       → "Nvidia owns the essential infrastructure for the AI revolution..."
// result.keyStrengths   → ["Dominant ~80%+ data center GPU share", "CUDA software moat", ...]
// result.keyRisks       → ["Customer concentration", "Cyclical demand exposure", ...]
// result.watchFor       → "Data center revenue growth rate. Deceleration below 30% YoY..."

Earnings Analysis

const result = await client.earningsAnalysis({ ticker: "MSFT" });
// result.verdict        → "strong_compounder" | "consistent" | "mixed" | "volatile" | "deteriorating"
// result.beatRate       → "4/5 quarters beat (80%)"
// result.revenueTrend   → "accelerating" | "stable" | "decelerating" | "declining"
// result.lastQuarterSummary → "..."

Insider Signal

const result = await client.insiderSignal({ ticker: "AAPL" });
// result.signal         → "strong_buy" | "buy" | "neutral" | "sell" | "strong_sell"
// result.confidence     → "high" | "medium" | "low"
// result.interpretation → "Two C-suite executives bought ~47k shares in March (positive alignment)..."

Valuation Snapshot

const result = await client.valuationSnapshot({ ticker: "GOOGL" });
// result.verdict        → "very_cheap" | "cheap" | "fair" | "expensive" | "very_expensive"
// result.metrics.peRatio, result.metrics.evEbitda, result.metrics.fcfYield, result.metrics.roe
// result.buyZone        → "Below $135. Current $172 is fair but not a bargain."

Bear vs. Bull

const result = await client.bearVsBull({ ticker: "TSLA" });
// result.verdict        → "bull_wins" | "slight_bull" | "too_close" | "slight_bear" | "bear_wins"
// result.bullCase       → [{ argument: "...", detail: "..." }, ...]   // 3 items
// result.bearCase       → [{ argument: "...", detail: "..." }, ...]   // 3 items
// result.keyDebate      → "Will autonomous driving become a real revenue line by 2027?"

Compare Stocks

Head-to-head comparison of 2–3 tickers with a recommendation by investor goal.

const result = await client.compareStocks({ tickers: ["NVDA", "AMD"] });
// result.winner         → "NVDA" | "AMD" | "tied"
// result.oneLiner       → "NVDA wins on margins and software moat; AMD better value."
// result.byTicker       → { NVDA: { strengths: [...], concerns: [...] }, AMD: {...} }
// result.ifYouValue     → { growth: "NVDA", value: "AMD", quality: "NVDA" }

Moat Analysis

Buffett-style competitive moat assessment.

const result = await client.moatAnalysis({ ticker: "KO" });
// result.moatRating     → "wide" | "narrow" | "none"
// result.moatSources    → [{ type: "brand", strength: "strong", evidence: "..." }, ...]
// result.durabilityRead → "Coca-Cola's brand moat is among the most durable in consumer staples..."
// result.threats        → ["Health-conscious consumer shift", "Private label competition", ...]

Text Extractor

Pull structured data out of raw text — no regex required.

const result = await client.textExtractor({
  text: "Contact Sarah at [e-mail] or [phone]. Budget: $12,500.",
  extractors: ["emails", "phone_numbers", "currencies"],
});
// result.extracted.emails       → ["[e-mail]"]
// result.extracted.phone_numbers → ["[phone]"]
// result.extracted.currencies   → ["$12,500"]

Token Counter

Never get surprised by context window costs again.

const result = await client.tokenCounter({
  text: longDocument,
  models: ["gpt-4o", "claude-3-5-sonnet", "gpt-3.5-turbo"],
});
// result.results["gpt-4o"].tokens            → 1842
// result.results["gpt-4o"].estimatedCost.input → 0.0000092 (USD)

Schema Generator

Describe your data in English, get back a schema.

const result = await client.schemaGenerator({
  description: "a SaaS user with name, email, plan tier, and usage limits",
  format: "typescript",
});
// result.schema → full TypeScript interface

CSV to JSON

Drop in CSV, get back typed JSON.

const result = await client.csvToJson({
  csv: rawCsvString,
  typeCast: true,   // "true" → true, "42" → 42, "" → null
});
// result.rows → [{ name: "Alice", age: 30, active: true }, ...]
// result.columnTypes → { name: "string", age: "number", active: "boolean" }

Markdown Converter

Clean up HTML for LLM consumption.

const result = await client.markdownConverter({
  content: scrapedHtml,
  from: "html",
  to: "markdown",
});
// result.output → clean Markdown without tags

URL Metadata

Enrich any link with context.

const result = await client.urlMetadata({ url: "https://example.com/article" });
// result.metadata.title       → "Article Title"
// result.metadata.description → "Meta description..."
// result.metadata.og          → { image: "...", type: "article" }

Other tools

// Natural language → cron expression
await client.cronBuilder({ description: "every weekday at 9am", timezone: "America/New_York" });

// Natural language → regex with code snippets
await client.regexBuilder({ description: "US phone numbers", testStrings: ["555-867-5309"] });

// Normalize messy US addresses to USPS format
await client.addressNormalizer({ address: "123 main st apt 4b, springfield, il 62701" });

// Generate color palettes from descriptions or hex colors
await client.colorPalette({ description: "calm fintech blue", count: 5 });

// Generate a full brand kit — colors, typography, CSS/Tailwind tokens
await client.brandKit({ name: "Solaris Health", industry: "healthcare", vibe: ["modern", "trustworthy"], format: "full" });

// Extract action items, decisions, and summary from meeting notes
await client.meetingActionItems({ notes: transcript, format: "full", participants: ["Sarah", "John"] });

// Analyze and improve an LLM prompt
await client.promptOptimizer({ prompt: "Summarize this and tell me the main points.", model: "gpt-4o", mode: "both" });

// Fetch a URL, strip boilerplate, return clean Markdown + AI summary
await client.webSummarizer({ url: "https://example.com/article", mode: "both", focus: "key arguments" });
// result.content           → clean Markdown
// result.summary.summary   → "2-4 sentence summary..."
// result.summary.keyPoints → ["point 1", "point 2", ...]

LangChain Integration

Use Agent Toolbelt tools directly in LangChain agents and chains.

import { AgentToolbelt } from "agent-toolbelt";
import { createLangChainTools } from "agent-toolbelt/langchain";
import { createReactAgent } from "@langchain/langgraph/prebuilt";
import { ChatOpenAI } from "@langchain/openai";

const client = new AgentToolbelt({ apiKey: process.env.AGENT_TOOLBELT_KEY! });
const tools = createLangChainTools(client);

const agent = createReactAgent({
  llm: new ChatOpenAI({ model: "gpt-4o" }),
  tools,
});

const result = await agent.invoke({
  messages: [{ role: "user", content: "Extract all emails and phone numbers from this text: ..." }],
});

Available LangChain tools

Tool name	Description
stock_thesis	Motley Fool-style investment thesis with live data
earnings_analysis	EPS beat/miss history + revenue trend read
insider_signal	Form 4 interpretation: meaningful signal or routine noise?
valuation_snapshot	P/E, P/S, EV/EBITDA, FCF yield → cheap/fair/expensive
bear_vs_bull	Steelmanned bull and bear cases with net verdict
compare_stocks	Head-to-head 2–3 ticker comparison with recommendation map
moat_analysis	Buffett-style competitive moat assessment with durability outlook
extract_from_text	Extract emails, URLs, phones, dates, currencies, addresses, names from text
count_tokens	Count tokens and estimate cost across LLM models
generate_schema	Generate JSON Schema / TypeScript / Zod from a description
csv_to_json	Convert CSV to typed JSON with auto type casting
convert_markdown	Convert HTML ↔ Markdown
fetch_url_metadata	Get title, description, OG tags, favicon from a URL
build_regex	Build and test regex patterns from natural language
build_cron	Convert schedule descriptions to cron expressions
normalize_address	Normalize US addresses to USPS format
generate_color_palette	Generate color palettes with WCAG scores and CSS variables
generate_brand_kit	Generate full brand kit — colors, typography, CSS/Tailwind tokens
strip_image_metadata	Strip EXIF/GPS/IPTC/XMP metadata from images for privacy
extract_meeting_action_items	Extract action items, decisions, and summary from meeting notes
optimize_prompt	Analyze and improve LLM prompts with scores and rewrite
compare_documents	Semantic diff between two document versions
extract_contract_clauses	Extract key clauses and flag risks from contracts
mock_api_response	Generate realistic mock data from a JSON Schema
pack_context_window	Select the best content subset that fits a token budget
audit_dependencies	Check npm/PyPI packages for known CVEs
summarize_web_page	Fetch a URL, extract clean Markdown, generate AI summary with key points

All tools

28 tools available (8 stock + 20 utility). Free tier included — paid tiers available for higher volume. See pricing.

Stock research ($0.05 per call)

Tool	Description
stock-thesis	Motley Fool-style investment thesis with verdict, strengths, risks
earnings-analysis	EPS beat/miss history + revenue trend over 5 quarters
insider-signal	Form 4 interpretation distinguishing real signal from routine
valuation-snapshot	P/E, P/S, EV/EBITDA, FCF yield → cheap/fair/expensive verdict
bear-vs-bull	Steelmanned bull and bear cases with net verdict
compare-stocks	Head-to-head 2–3 ticker comparison with winner
moat-analysis	Buffett-style competitive moat with durability rating

Utility tools

Tool	Description
schema-generator	JSON Schema / TypeScript / Zod from natural language
text-extractor	Extract emails, URLs, phones, dates, currencies from text
token-counter	Token counts and cost estimates across LLM models
csv-to-json	CSV to typed JSON with auto type detection
markdown-converter	HTML ↔ Markdown conversion
url-metadata	Title, description, OG tags, favicon from a URL
regex-builder	Natural language → tested regex with code snippets
cron-builder	Natural language → cron expression with next run times
address-normalizer	Normalize US addresses to USPS format
color-palette	Color palettes with WCAG scores and CSS variables
brand-kit	Full brand kit — colors, typography, CSS/Tailwind tokens
image-metadata-stripper	Strip EXIF/GPS/IPTC/XMP metadata from images
meeting-action-items	Extract action items and decisions from meeting notes
prompt-optimizer	Analyze and improve LLM prompts
document-comparator	Semantic diff between two document versions
contract-clause-extractor	Extract clauses and flag risks from contracts
api-response-mocker	Generate realistic mock data from a JSON Schema
context-window-packer	Pack content into a token budget optimally
dependency-auditor	Audit npm/PyPI packages for CVEs
web-summarizer	Fetch a URL, extract clean Markdown, generate AI summary with key points

Going to production

When the agent moves out of dev, a new set of questions shows up. What did it call last night? What arguments did it pass? Who approved the destructive one?

Cordon is an MCP gateway that sits in front of servers like this one. Point your client at Cordon instead of directly at Agent Toolbelt; Cordon forwards every call through and adds:

• A real-time audit log of every tool invocation (name, arguments, response, latency)
• Per-API-key policy: which tools each caller can use, under what conditions
• Slack-based human approvals for tool calls you've flagged as high-risk

From the agent's perspective nothing changes — same tools, same schemas. Free tier covers 250 events/month.

License

MIT