Ravi Ramanujan
Address parsing and accuracy benchmarks for sthan.io
Ravi writes about turning messy address text into clean fields: typos, missing ZIP codes, abbreviations and words in the wrong order, and how often a parser gets it right on input like that.
His accuracy numbers come from a benchmark of 26,943 corrupted versions of real US addresses, and the articles report where the parser falls short as well as where it works.
- 5 articles
- 44 minutes of reading
- 26,943 addresses tested
- Latest: Oct 16, 2026
What Ravi covers
- How parsing works AI Address Parsing: Messy US Address Text to Clean Datawhat goes wrong with freeform input, with before-and-after examples
- Choosing a parser Address Parser API: What It Returns, Comparedlive responses on messy input and five parser APIs with prices
- How accurate it is How Accurate Is US Address Parsing on Messy Input?recovery, wrong-answer rate, and refusing addresses that do not exist
- Three ways to call it How to Parse Messy US Addresses with the API, MCP, and CLIcopy-paste examples for the REST API, AI assistants and the terminal
Where to start
A reading order for someone with a column of messy addresses.
-
AI Address Parsing 10 minStart here to see the kinds of mess a parser has to handle.
-
See what you get back and how the options compare on price.
-
Check the numbers before you trust a parser with your data.
-
Pick the way to call it that fits your workflow.
Questions these articles answer
Each question links to the answer inside the article.
The basics
- What is freeform address input? AI address parsing
- What is an address parser API? Parser API compared
- Can an API parse an address from a single text string? AI address parsing
- Does an address parser verify the address? Parser API compared
- What is the difference between address parsing and address verification? AI address parsing
Accuracy
- How accurate is address parsing on messy input? Accuracy benchmark
- What is a silent substitution in address parsing, and how often does it happen? Accuracy benchmark
- Can I set an automatic accept threshold? Accuracy benchmark
- How does sthan.io handle typos and misspellings in addresses? AI address parsing
- What happens when part of the address is missing? AI address parsing
Using it
- Can an AI assistant parse addresses directly? API, MCP and CLI
- Is there a command-line tool for parsing addresses? API, MCP and CLI
- Can a parser API pull addresses out of free text? Parser API compared
- How much does an address parser API cost? Parser API compared
How the numbers were made
- Benchmark
- 26,943 corrupted versions of real US addresses, drawn from a 200,000-address set screened so only confirmable addresses anchor the score
- Reported
- Recovery, wrong answers, how often the parser says it is unsure, and refusals of fake addresses
- Compared
- Five parser APIs and their prices, checked Oct 5 to 6, 2026
- Ways to call it
- REST API, MCP server for AI assistants, and a command-line tool
Related on sthan.io
Terms explained
All articles by Ravi
Address Validation MCP Servers for AI Agents, Compared
Eight address validation MCP servers for AI agents, updated October 28, 2026: tools, transport, sign-up, cost, registry status, and live tool calls shown.
Address Parser API: What It Returns, Compared
What a US address parser API returns, shown with live responses on messy input, the accuracy on 26,943 corrupted real addresses, and five parser APIs and their prices side by side.
How Accurate Is US Address Parsing on Messy Input?
26,943 corrupted versions of real US addresses: recovery, wrong-answer rate, disclosure, and how the parser handles addresses that do not exist.
How to Parse Messy US Addresses with the API, MCP, and CLI
Parse freeform US addresses three ways, with copy-paste examples for the REST API, the MCP server for AI assistants, and the CLI.
AI Address Parsing: Messy US Address Text to Clean Data
How sthan.io handles typos, missing ZIP codes, abbreviations, and out-of-order components in freeform address text.