Query over HTTP
Pull a nested value with a dot-path, index into an array, or filter an array of objects with eq / ne / gt / lt / like. Project one field out of the matches, or cast the result to text, number or bool.
One document becomes an endpoint your apps read, your scripts query and your AI tools edit, with a schema check on every write and edge caching on every read.
Trial2 days, no card required · everything unlocked · upgrade any time
No connection string, no ORM, no migrations. Just a document you address by URL and shape with query-string operators.
Pull a nested value with a dot-path, index into an array, or filter an array of objects with eq / ne / gt / lt / like. Project one field out of the matches, or cast the result to text, number or bool.
Save a blob, get a URL. No server to run, no schema to define.
Private behind an API key, or flip it public for a no-auth read. Your call, per silo.
Pin a silo to Germany, Singapore or the United States so reads are served close to whoever is calling. Basic and Pro can use all three.
Every plan has a per-minute burst limit and a request cap, and every response carries live *-Req-* headers. On the trial and every paid plan, a dashboard also breaks requests down per silo by day, week or month.
1,284 requests · 3 silos · last 7 days
Reads are served from a cache at the edge: roughly 30 minutes for a whole silo, 5 for a query result. A write to a silo clears its cache straight away, so hot endpoints stay fast without ever going stale.
One endpoint on api.jsonsilo.com/<uuid>. The query string picks one of four branches, checked top to bottom, and the first one that matches runs. No schema, nothing to set up.
Every read is the same GET. Which branch runs depends only on which params you send.
q / value / opFilter a top-level array. Add path= to project one field out of every match.
idxOne element of a top-level array, by position.
pathDot-path into the root document. A numeric segment indexes an array.
no paramsNothing to match, so the whole silo comes back.
Six comparisons, one per condition.
A three-item catalog, stored exactly as you posted it. Every example here reads from it.
[{ "name": "Notebook", "price": 6, "inStock": true,"tags": [{ "label": "paper" }, { "label": "sale" }] },{ "name": "Pen", "price": 2, "inStock": true,"tags": [{ "label": "ink" }] },{ "name": "Backpack", "price": 45, "inStock": false,"tags": [{ "label": "bags" }, { "label": "sale" }] }]
A filter is not limited to the element's own fields. Dots walk down into it.
Dot-separated keys. A numeric segment indexes an array.
Cast the value on the way out. Raw JSON by default.
Repeat the trio to AND them. Send the same number of each.
?q=price&value=5&op=gt
&q=price&value=50&op=lt
&path=name
["Notebook", "Backpack"]
Lock a silo to a JSON Schema so a bad write never reaches the apps reading it, then let Claude Code, Cursor or VS Code read and edit your silos over MCP.
Attach a JSON Schema to a silo and lock it. Any write that doesn't match is rejected, and the stored data stays exactly as it was.
draft-04, draft-07, 2019-09, 2020-12.[
{ "name": "Notebook", "price": 6, … },
{ "name": "Pen", "price": "2", … },
…
]/1/price · Instance type "string" is invalid. Expected "number".
Stored data unchanged. Apps keep reading the last good version.
A remote MCP endpoint for your silos. Create an MCP key, paste one snippet, and your AI tool can list, read, query and edit silos in plain language.
patch_silo applies JSON Patch atomically and still answers to the schema lock.Both are included in the trial and every paid plan. Read the schema lock and MCP docs.
Every plan includes the query engine, public and private silos, and edge caching. Start with a full-access trial, then pay for capacity (silos, size and daily requests) — not seats.
Every feature unlocked, no card required.
For growing projects.
For serious workloads.
Need more than Pro? Talk to us.
Still stuck? The docs go deeper on every endpoint and operator.
A silo is one JSON file with its own web address. You save it once, then read all of it or just the parts you need over HTTP.
Only if you choose to make it public. New silos are private, so no one can read them without your key.
You send a key with each request. There's a read key for fetching data and a separate key for creating or changing silos from your own code.
Yes. You can reach a value deep inside the file, pick one item out of a list, or filter a list down to the items you want.
Each plan has a per-minute burst limit. The trial is capped at 1,000 requests total over its 2 days; paid plans get a daily limit instead, from 1 million a day on Basic up to 10 million on Pro. Every response tells you how many requests you have left.
There's nothing to set up or connect to. You just read a file from a URL. It's made for config, content and small datasets, not for heavy transactional apps.
Germany, Singapore and the United States. The trial and every paid plan can route through any of the three.
Yes. Attach a JSON Schema to a silo and lock it, and any write that doesn't match is rejected with an error naming the exact field. The stored data stays as it was. You can generate the schema from your current data instead of writing it by hand.
Yes, over MCP. Create an MCP key in the console and add JSONsilo to Claude Code, Cursor, VS Code or any client that supports Streamable HTTP. A read-only key lets the assistant list, read and query silos; a read & write key also lets it create, update, patch and delete them.
Yes. Full reads are cached for about 30 minutes and query results for about 5, so busy endpoints stay fast. Saving a change clears the cache right away.
Sign up, paste some JSON, and copy the URL into your code. The free trial is enough to ship something real.