APIShed
A modular HTTP API server. Each module owns its own versioned route prefix.
The math LLMs get wrong — as a deterministic API and MCP server.
Checksums, date arithmetic, exact character counts, token counts — a language model guesses at these instead of computing them. apished runs the actual algorithm and gives you a real answer, over plain HTTP or as MCP tools an agent can call directly.
GPT-4's accuracy on timezone questions vs. date-difference questions, on the "Test of Time" benchmark.
Date math, done right — /v1/datemath
The canonical LLM answer to "how many r's in strawberry" — tokenization groups letters into tokens, so the model never actually counts.
Character counts you can trust — /v1/text
Ask an LLM for a SHA-256 hash and it will confidently invent a real-looking hex string instead of computing one.
Real checksums, not guesses — /v1/hash
Rules
- Free to use, no API key or signup required.
- Rate limited to 1 request/second per IP per endpoint, across both the REST API and MCP tools — calling several different endpoints in the same second is fine, only hammering the same one is throttled. Exceeding it returns HTTP 429 (or an MCP tool error).
- No request or response payloads are logged. Only an aggregate call count per endpoint is kept — visible live at /stats — with no IP addresses or other client-identifying data stored anywhere.
MCP
/mcpEvery module below, exposed as 29 MCP tools over Streamable HTTP for MCP clients like Claude Desktop, Claude Code, Cursor, and ChatGPT — same logic as the REST endpoints, no duplication. See /mcp-setup for the full tool list and client-by-client setup instructions.
Modules
Every module's routes, grouped below. For the full interactive reference — request/response schemas, try-it-out — see /docs (openapi.json).
Evaluates Reverse Polish Notation expressions — arithmetic, trig, logs, bitwise ops, and constants.
Validates Polish PESEL, NIP, and REGON checksums — PESEL also decodes birth date and sex.
Validates Luhn, ISBN, EAN/UPC, IBAN, EU VAT, and SWIFT/BIC checksums and formats.
Computes MD5/SHA1/SHA256/SHA512/CRC32 digests, with optional base64/hex input.
Character/word counting, reversal, palindrome checks, indexing, and sorting — Unicode-safe.
Calendar date arithmetic — differences, business days, age, day-of-week, add/subtract.
Regex matching, email/URL format checks, and JSON Schema validation.
Counts tokens under OpenAI-compatible (tiktoken) encodings, by model or encoding name.