Small APIs that buy you big time.

MCP tools that give AI agents certainty. Compute instead of guess—no hesitation, no waiting, safe connection to external knowledge.

Get Started View Docs

{
  "mcpServers": {
    "thousand-api": {
      "url": "https://mcp.thousand-api.com/mcp",
      "headers": {
        "x-api-key": "your-api-key"
      }
    }
  }
}

Cursor uses Streamable HTTP directly. Claude Desktop uses mcp-remote via npx (Node.js required).

Certainty

Guarantee correctness through computation, not inference

Replace areas where LLMs fail at mental math or recall—crypto, compound interest, boundary checks—with deterministic tool calls.

Token efficiency

Compress long DIY and retries into one tool call

Cut wasted reasoning tokens and retries so agent responses stay faster and cheaper.

Agent-native

MCP-ready—configure and go

No SDK or wrapper required. Connect Claude, Cursor, and other MCP clients directly.

Why you can trust it

Tool counts and verified results that back up certainty.

84tools

7categories

70+verified evaluations

Where AI often fails

The same prompts, tried with a typical AI chat and Thousand API (via MCP).

🔐 SHA-256 hash generation

Typical AI chat

aa69305e7a9… (a value that was not actually correct)

Thousand API

Returns the correct SHA-256 hash immediately

🌐 DNS lookup (github.com MX)

Typical AI chat

aspmx.l.google.com (stale training-data answer)

Thousand API

outlook.com (the live MX record)

🔗 Short URL resolution

Typical AI chat

Confidently invents a URL that does not exist

Thousand API

Follows HTTP redirects and returns the final URL

📅 Business days (10 days after 6/27)

Typical AI chat

July 28 (off-by-one on inclusive start)

Thousand API

July 29 (exact per the rule)

📚 ISBN-13 validation

Typical AI chat

Marked every valid ISBN-13 as invalid

Thousand API

Validates correctly via check digit

🔑 TOTP generation

Typical AI chat

Refused / gave up computing TOTP for an unpublished secret

Thousand API

Generates deterministic TOTP per RFC 6238

💰 Loan payment (equal principal & interest)

Typical AI chat

Wrong monthly payment (answered ¥42,696)

Thousand API

Correct monthly payment: ¥42,749

🏷️ Domain format check (example.123)

Typical AI chat

Said a numeric-only TLD is still valid

Thousand API

Correctly rejects the invalid TLD

July 2026 / comparison vs ChatGPT free tier

Certainty

ChatGPT vs Thousand API

Why certainty matters. Three verified logs—same prompts, real answers.

TOTP generation

Refuses to computesecurity.generate_totp

Prompt

For an unpublished random Base32 secret and a UNIX timestamp, compute the TOTP codes (6-digit and 8-digit) and expires_in per RFC 6238. Answer with computed values only—do not guess.

ChatGPT (no MCP)

This calculation involves a cryptographic hash (HMAC-SHA1) and binary operations, so producing exact values requires actual computation (code execution). Hand calculation or guessing cannot guarantee RFC 6238–compliant results.

Produced no numeric result—abandoned the answer

Thousand API (via MCP)

6-digit: 410078 / 8-digit: 19410078 / expires_in: 13

Exact match with precomputed ground truth

Correct answer: 6-digit 410078, 8-digit 19410078, expires_in 13

Why the difference: Public RFC test vectors can be recalled from training data, but unpublished inputs lead the model to refuse computation. HMAC-based TOTP is a classic case where LLM inference cannot substitute for real crypto.

Loan repayment schedule

Confidently wrong numbersutility.calc_loan

Prompt

Principal ¥1,000,000, annual rate 2.475%, 24 months, equal principal-and-interest payments. Compute monthly payment, total payment, total interest, and month-1 / final-month breakdowns. Round to yen; clear remainder in the final month.

ChatGPT (no MCP)

Item	ChatGPT	Correct
Monthly payment	¥42,696	¥42,749
Total payment	¥1,026,023	¥1,025,984
Final-month payment	¥44,015	¥42,757

Plausible-looking figures, but the monthly payment is already ¥53 off—every later line cascades from that error.

Thousand API (via MCP)

Monthly payment: ¥42,749 / Total payment: ¥1,025,984 / Final month: ¥42,757 (balance clears to ¥0)

Exact match with precomputed ground truth

Correct answer: Monthly ¥42,749 / Total ¥1,025,984 / Final month ¥42,757

Why the difference: The model does not refuse—it returns wrong numbers with confidence. Without an independent check, money math errors are the most dangerous failure mode in production.

Domain format validation

Plausible misclassificationnetwork.check_domain_format

Prompt

For four domains (Japanese-domain.jp / -invalid.com / example.123 / localhost), decide whether each is valid per RFC 1035/1123. For the Japanese domain, also return the Punycode form.

ChatGPT (no MCP)

Item	ChatGPT	Correct
example.123	Valid (numeric-only TLD OK in form)	Invalid (numeric-only TLD)
localhost	Valid (RFC 6761 special-use)	Invalid (no TLD)
Japanese-domain.jp Punycode	Wrong once, then self-corrected	xn--eckwd4c7c5976acvb2w6i.jp

Cites plausible reasons (even RFC numbers) while misjudging edge cases—hard to spot without domain expertise.

Thousand API (via MCP)

All four answers correct (including Punycode)

Rule-based checks stay consistent on boundary cases

Correct answer: example.123 and localhost are invalid. Japanese-domain Punycode: xn--eckwd4c7c5976acvb2w6i.jp

Why the difference: Edge cases—numeric-only TLDs, single-label names—need deterministic rule evaluation, not narrative justification.

Verified July 2026 / compared with ChatGPT free tier

ChatGPT answers reflect the verification date. Future model updates may change responses.

Bring this certainty to your AI agents

Copy the MCP config—agents return deterministic results instead of guesses.

Get Started View Docs

MCP Tools

Reliable utility MCP tools for AI agents and developers

84 MCP tools available

Date & Calendar

16 tools

On “10 business days later,” typical LLMs often miscount the start date. Thousand API calculates exactly to the specification.

Text & String

12 tools

Code generation, text processing, and Japanese conversion with reliable, token-efficient APIs

Convert & Format

13 tools

Format conversions for AI-generated data with zero tokens

Data Processing

10 tools

Extract values from large JSON, validate schemas, compute statistics

Network & Infra

11 tools

Typical LLMs pretend to know DNS records and URL contents. Thousand API actually fetches and verifies them.

Security & Auth

8 tools

SHA-256 and HMAC invite LLMs to invent plausible fake values. Thousand API performs real cryptographic computation.

Developer Utilities

14 tools

Distance, coordinates, color tools, and version comparison with deterministic results

Tool names, endpoints, parameters, and response examples are in the

MCP tool index →

Save LLM tokens and ensure accurate results

When AI agents use tools, they get accurate results with fewer tokens than having the LLM process everything itself.

Two approaches to reducing tokens

① Use tools to reduce output tokens

e.g. 95% reduction with HTML→text extraction

② Turn off unused tools to reduce tool-definition tokens

Configure per category or per tool from the dashboard
All 84 tools → 20 tools cuts tool-definition tokens by about 76%

Common filtering presets:

Developer set (6 tools) → Security + Data processing
Document set (5 tools) → Text + Format conversion
Research set (5 tools) → Network

94%

Token reduction

HTML to text

64%

Token reduction

Postal code lookup

67%

Token reduction

Compound interest

60%

Token reduction

HMAC verification

Without tools

Input

8,200 TOKEN

Output

1,100 TOKEN

Total: 9,300 TOKEN

With tools

Input

380 TOKEN

Output

160 TOKEN

Total: 540 TOKEN

↑ 94% saved

See all scenarios →

Use cases

Hand agents MCP tools for work that breaks when left to inference

AI assistant

Scheduling for AI assistants

Meeting times and reminders drift when agents infer business days, time zones, or cron schedules.

Related tools

Holiday & Business Day
Cron Parser

Learn more

AI marketing

Prep work for AI writers and marketers

Outlines and competitor research suffer when meta tags and post timing are guessed.

Related tools

Page Meta / OGP
QR Code Generator
Holiday & Business Day

Learn more

Cursor / Claude dev

Precise utilities for AI engineers

JSON, cron, and URLs in generated code introduce subtle spec mistakes.

Related tools

JSON Validator
Cron Parser
URL Redirect Resolver

Learn more

Chatbot

External sensors for chatbots

Bots lack live locale, link targets, and final URLs for short links.

Related tools

IP Geolocation
Page Meta / OGP
QR Code Generator
URL Redirect Resolver

Learn more

View all use cases

Works as an MCP server

Connect directly to Claude or Cursor. Copy the config and your AI agent can call Thousand API tools right away.

Supported clients

Claude
Cursor
Other MCP-compatible tools

{
  "mcpServers": {
    "thousand-api": {
      "url": "https://mcp.thousand-api.com/mcp",
      "headers": {
        "x-api-key": "your-api-key"
      }
    }
  }
}

Cursor uses Streamable HTTP directly. Claude Desktop uses mcp-remote via npx (Node.js required).

Pricing

84 MCP tools free for up to 1,000 requests/month. Upgrade as you scale.

Free

Trial and light personal use

MCP server connection
REST API access
API keys1
Monthly requests1,000
Email support
Priority support

Get Started

Recommended

Starter

¥480/mo

Everyday personal use

MCP server connection
REST API access
API keys5
Monthly requests10,000
Email support
Priority support

Start Starter

Pro

¥1,980/mo

Power users and small teams

MCP server connection
REST API access
API keys20
Monthly requests100,000
Email support
Priority support

Start Pro

Get Started