JSONsilo
your JSON, at the center

Your JSON, wired to everything.

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

Your appfetch(url)Scripts?q=…&op=eqClaude Codepatch_siloCursorupdate_siloGermanySingaporeUnited StatesSchema lockMCPQueryEdge cache{ }product-catalog
  • Reads, edge-cached
  • Writes, schema-checked
  • Live edge region
  • Instant endpointsSave JSON, get a URL
  • Query over HTTPPaths, filters, casts
  • Schema lockBad writes get a 422
  • 3 edge regionsGermany, Singapore, United States
  • MCP for AI toolsClaude Code, Cursor, VS Code
The platform

Everything a small dataset needs, and nothing it doesn't

No connection string, no ORM, no migrations. Just a document you address by URL and shape with query-string operators.

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.

GETapi.jsonsilo.com/60f97483-…?path=0.name
"Notebook" · value at a dot-path

Instant JSON endpoints

Save a blob, get a URL. No server to run, no schema to define.

POST{ "name": "Notebook", … }201api.jsonsilo.com/60f97483-…GET{ "name": "Notebook", … }

Public or private

Private behind an API key, or flip it public for a no-auth read. Your call, per silo.

PrivateX-SILO-KEY
Public/public/<uuid>

Edge regions

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.

Germanyapi.jsonsilo.com
Singaporesg-01-api.jsonsilo.com
United Statesusc-01-api.jsonsilo.com

Request analytics & limits

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.

dayweekmonth

1,284 requests · 3 silos · last 7 days

*-Req-Limit*-Req-Remaining*-Req-Reset

Built-in caching

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.

HITserved from the edgeage 214s
Whole silo
~30 min
Query · index · path
~5 min
Metadata
~6 hr
On write
cleared
Query engine

Ask for exactly the slice you need

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.

Four branches, first match wins

Every read is the same GET. Which branch runs depends only on which params you send.

GETapi.jsonsilo.com/60f97483-…X-SILO-KEY
  1. q / value / op

    Filter a top-level array. Add path= to project one field out of every match.

    ?q=price&value=10&op=lt&path=name["Notebook", "Pen"]
  2. idx

    One element of a top-level array, by position.

    ?idx=1{ "name": "Pen", "price": 2, … }
  3. path

    Dot-path into the root document. A numeric segment indexes an array.

    ?path=0.tags.1.label"sale"
  4. no params

    Nothing to match, so the whole silo comes back.

    (no query string)[ { "name": "Notebook", … }, … ]

op=

Six comparisons, one per condition.

eq
equals
ne
not equal
gt
greater than
lt
less than
lk
pattern, % wildcard
il
pattern, any case
?q=name&value=Pen&op=eq&path=price[2]

The document

A three-item catalog, stored exactly as you posted it. Every example here reads from it.

silo.json
[
{ "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" }] }
]

q= reaches nested data

A filter is not limited to the element's own fields. Dots walk down into it.

name
a plain field on the element
profile.city
one dot walks a nested object
tags..label
two dots walk a nested array and match any item
?q=tags..label&value=sale&op=eq&path=name["Notebook", "Backpack"]

path=

Dot-separated keys. A numeric segment indexes an array.

0.nameprofile.address.citytags.0

output=

Cast the value on the way out. Raw JSON by default.

jsonjsonbtextintfloatbool
?idx=1&path=price&output=int2

Stack conditions

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"]

Guardrails and agents

Keep the shape. Hand the edits to your tools.

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.

Schema lock

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.

  • Generate it from your data. Loose allows new fields; strict rejects unknown ones.
  • Errors name the field. A failing write gets a 422 with the JSON Pointer path that broke.
  • Scripts can't loosen it. Manage keys have no schema routes; only the console can change the lock.
  • JSON Schema draft-04, draft-07, 2019-09, 2020-12.
product-catalogSchema60f97483-…
Generate from data
LooseStrict
PATCH/manage/60f97483-…
[
  { "name": "Notebook", "price": 6, … },
  { "name": "Pen", "price": "2", … },
  …
]
422Silo data does not match its locked schema.

/1/price · Instance type "string" is invalid. Expected "number".

Stored data unchanged. Apps keep reading the last good version.

MCP server

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.

  • Read-only or read & write. A read-only key never even sees the write tools.
  • Safe edits. patch_silo applies JSON Patch atomically and still answers to the schema lock.
  • Keys you control. Shown once, stored as a hash, revoked instantly.
  • Works with any client that speaks Streamable HTTP with custom headers.
Terminal
claude mcp add --transport http jsonsilo \
  https://mcp.jsonsilo.com/mcp \
  --header "Authorization: Bearer jsmcp_YOUR_KEY"
Read-only key
  • list_silos
  • get_silo_meta
  • read_silo
  • query_silo
  • get_silo_schema
Read & write key adds
  • create_silo
  • update_silo
  • patch_silo
  • delete_silo

Both are included in the trial and every paid plan. Read the schema lock and MCP docs.

Pricing

Try it free, scale when you need to

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.

Trial

No card
Freefor 2 days

Every feature unlocked, no card required.

  • 6 silos
  • 5 MB per silo
  • 3 regions
  • 1,000 requests total
  • Public & private silos
  • Query engine
  • Schema lock
  • MCP for AI tools
  • Request analytics
Start free trial

Basic

$9/month

For growing projects.

  • 6 silos
  • 5 MB per silo
  • 3 regions
  • 1M requests / day
  • API keys & manage keys
  • Schema lock
  • MCP for AI tools
  • Request analytics
Choose Basic

Pro

Most popular
$25/month

For serious workloads.

  • 111 silos
  • 10 MB per silo
  • 3 regions
  • 10M requests / day
  • Priority edge routing
  • Everything in Basic
Choose Pro

Need more than Pro? Talk to us.

FAQ

Questions, answered

Still stuck? The docs go deeper on every endpoint and operator.

What is a silo?

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.

Is my data public?

Only if you choose to make it public. New silos are private, so no one can read them without your key.

How do I authenticate?

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.

Can I query nested JSON?

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.

What are the rate limits?

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.

How is this different from a database?

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.

Which regions can I use?

Germany, Singapore and the United States. The trial and every paid plan can route through any of the three.

Can I stop a bad write from breaking my app?

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.

Can AI tools read and edit my silos?

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.

Do reads get cached?

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.

JSONsiloJSON in, URL out

Create your first silo in about two minutes

Sign up, paste some JSON, and copy the URL into your code. The free trial is enough to ship something real.