Write MongoDB aggregation queries in JavaScript. A strict JS subset that compiles to MQL JSON — like SQL but for MongoDB, using the syntax you already know.
import { jsmql } from "@koresar/jsmql";
// Filter — for db.coll.find(filter). No `;` at top level.
const age = 18;
let filter = jsmql`$.age > ${age} && $.status === "active"`
// → { age: { $gt: 18 }, status: "active" } ← index-friendly query doc
// Pipeline — for db.coll.aggregate(pipeline). Any `;` flips to stage mode.
// Snapshot one user, then pivot the stream onto their 5 most-recent orders.
let pipeline = jsmql`
$$ = $$.filter(u => u.email === "me@example.com").slice(0, 1);
let userId = $._id;
$$ = $$$$.archive.orders
.filter(o => o.userId === userId)
.toSorted((a, b) => a.placedAt - b.placedAt)
.toReversed()
.slice(0, 5);
`;
// → [
// { $match: { email: "me@example.com" } },
// { $limit: 1 },
// { $set: { "__jsmql.userId": "$_id" } },
// { $lookup: { from: { db: "archive", coll: "orders" }, let: { userId: "$__jsmql.userId" }, pipeline: [
// { $match: { $expr: { $eq: ["$userId", "$$userId"] } } },
// { $sort: { placedAt: -1 } },
// { $limit: 5 },
// ], as: "__jsmql.__lookup1" } },
// { $unwind: "$__jsmql.__lookup1" },
// { $replaceWith: "$__jsmql.__lookup1" },
// { $unset: "__jsmql" }
// ]
// Raw expression — for inside a stage body, or db.coll.updateOne(filter, update).
let expr = jsmql.expr(($) => $.items.map((i) => i.price * i.qty).reduce((a, x) => a + x, 0))
// → { $reduce: { input: { $map: { input: "$items", as: "i",
// in: { $multiply: ["$$i.price", "$$i.qty"] } } },
// initialValue: 0, in: { $add: ["$$value", "$$this"] } } }MongoDB 8.0 deprecated server-side JavaScript via $function, $accumulator, and $where. The JSMQL is the replacement: native MQL, no --noscripting issues, index-friendly, IDE-aware, testable as plain JS.
npm install @koresar/jsmqlESM + CJS, runs in browsers, zero dependencies. Works with Node 14+, Deno, and Bun.
import "@koresar/jsmql/ops"; // ambient $-prefixed globals — autocomplete for 182 MQL ops & every stage
import { jsmql } from "@koresar/jsmql";
// Arrow form — your prettier/oxfmt handles formatting.
// No `;` at top level → query Filter (the doc db.coll.find(filter) takes).
jsmql(($) => $.email === $.email.trim().toLowerCase().endsWith("@flash-payments.com"))
// → {"$expr":{"$eq":["$email",{"$eq":[{"$substrCP":[{"$toLower":{"$trim":{"input":"$email"}}},{"$subtract":[{"$strLenCP":{"$toLower":{"$trim":{"input":"$email"}}}},{"$strLenCP":"@flash-payments.com"}]},{"$strLenCP":"@flash-payments.com"}]},"@flash-payments.com"]}]}}
// Pipelines — any `;` flips to stage mode (the array db.coll.aggregate(pipeline) takes).
jsmql(($) => {
$match($.age >= 18 && $.region === "AU"); // → query doc, indexes still work
$group({ _id: $.shopId, total: { $sum: $.amount } });
$sort({ total: -1 });
});
// → [{ "$match": { "age": { "$gte": 18 }, "region": "AU" } }, { "$group": { "_id": "$shopId", "total": { "$sum": "$amount" } } }, { "$sort": { "total": -1 } }]
// Use `?.` where a field might be null — you get `$ifNull` guards exactly there:
jsmql('[...$.mods, ...$.room?.mods, "root"].includes($.userId)')
// { "$expr": { "$in": ["$userId", { "$concatArrays": ["$mods", { "$ifNull": ["$room.mods", []] }, ["root"]] }] } }
// `new Date(...)` with literal args folds to a real JS Date — index-friendly query doc:
jsmql(`$.method === "postalDelivery" && $.createdAt >= new Date("2026-01-01")`)
// → { method: "postalDelivery", createdAt: { $gte: <Date 2026-01-01> } }
// `new Date()` and `new Date($.field)` still need server-time evaluation and ride in $expr.
// Template-tag — interpolate runtime literals from outer scope
const ids = [1, 2, 3];
jsmql`$.status === "open" && $.id in ${ids}`
// → { "status": "open", "$expr": { "$in": ["$id", [1, 2, 3]] } }
// jsmql.compile — parse once, bind many. Output stays index-friendly.
const eligible = jsmql.compile(({ minAge, region }, $) => {
$match($.age >= minAge && $.region === region);
$project({ age: 1, email: 1, address: 1 });
});
eligible({ minAge: 21, region: "AU" });
// → [{"$match":{"age":{"$gte":21},"region":"AU"}},{"$project":{"age":1,"email":1,"address":1}}]
// JS-natural `=`, `+=`, `delete` compile to coalesced $set / $unset
jsmql(($) => {
$.score += 1;
delete $.tempToken;
$.status = "done";
});
// → [{ "$set": { "score": { "$add": ["$score", 1] } } }, { "$unset": "tempToken" }, { "$set": { "status": "done" } }]
// Assigning to bare `$` replaces the whole document — lowers to $replaceWith
jsmql(`$match($.profile != null); $ = $.profile; $ = { ...$, score: $.points * 1.1 }`);
// → [
// { "$match": { "profile": { "$ne": null } } },
// { "$replaceWith": "$profile" },
// { "$replaceWith": { "$mergeObjects": ["$$ROOT", { "score": { "$multiply": ["$points", 1.1] } }] } }
// ]
// Multi-facet aggregation — every value a `$$.filter(...)` lowers to one $facet stage
jsmql(`$ = {
topByScore: $$.filter(o => { $sort({ score: -1 }); $limit(10); }),
recent: $$.filter(o => o.createdAt >= "2026-01-01"),
byStatus: $$.filter(o => { $group({ _id: o.status, n: $sum(1) }); })
}`);
// → [{ "$facet": {
// "topByScore": [{ "$sort": { "score": -1 } }, { "$limit": 10 }],
// "recent": [{ "$match": { "createdAt": { "$gte": "2026-01-01" } } }],
// "byStatus": [{ "$group": { "_id": "$status", "n": { "$sum": 1 } } }]
// } }]
// export contact details of active users (`$$ = [$$.reduce(… => acc.concat(…), [])]`)
// to a different collection in the 'exports' database
jsmql(`
$group({ _id: $.userId, revenue: $sum($.total), orders: $sum(1) });
$$ = $$.toSorted((a, b) => b.revenue - a.revenue).slice(0, 10);
`);
// [
// { "$group": { "_id": "$userId", "revenue": { "$sum": "$total" }, "orders": { "$sum": 1 } } },
// { "$sort": { "revenue": -1 } },
// { "$limit": 10 }
// ]
// `jsmql()` returns an UpdateFilter as a pipeline, to avoid common footgun of wiping out the whole collection.
db.users.updateMany({}, jsmql(($) => $.name = $.name.toUpperCase()))
// → [{ "$set": { "name": { "$toUpper": "$name" } } }] -> will upper-case all names in the collection
// `jsmql.expr()` returns a partial MQL JSON. Won't protect from the same footgun.
db.users.updateMany({}, jsmql.expr(($) => $.name = $.name.toUpperCase()))
// → { "$set": { "name": { "$toUpper": "$name" } } } -> will WIPE OUT all names in the collection
// Strict-shape entry points — throw if the input would produce the wrong shape.
// Use these when the call site demands a specific shape and a silent
// mis-dispatch would be a footgun.
db.users.find(jsmql.filter("$.age > 18")); // throws on Pipeline-shaped input
db.users.aggregate(jsmql.pipeline("$match($.age > 18); $sort({ age: 1 })")); // throws on bare expressions
db.users.updateOne({ _id: 1 }, jsmql.update("$.name = $.name.toUpperCase()"));
// update() additionally rejects any stage outside MongoDB's update-pipeline
// whitelist ($addFields, $project, $replaceRoot, $replaceWith, $set, $unset),
// so a misplaced `$match` is caught at compile time instead of at the server.
// Raw expression — for embedding inside a hand-written stage body
const stage = { $addFields: { discount: jsmql.expr(($) => $.price * (1 - $.loyalty.multiplier)) } }
// → { $addFields: { discount: { $multiply: ["$price", { $subtract: [1, "$loyalty.multiplier"] }] } } }
// Escape hatch — call any MongoDB operator as a function - $dateTrunc in this case
jsmql.expr(($) => $set({ createdAtWeek: $dateTrunc({ date: $.createdAt, unit: "week" }) }))
// → { $set: { "createdAtWeek": { "$dateTrunc": { "date": "$createdAt", "unit": "week" } } } }
jsmql(($) => $.age = 18); // generates a pipeline, to make sure you can use this in updateOne(), updateMany(), etc
// → [{ "$set": { "age": 18 } }]
jsmql.expr(($) => $.age = 18); // generates an partial expression, to use within OTHER aggregation or filter expressions
// → { "$set": { "age": 18 }
// Validate without throwing — every error carries { message, pos, code }
jsmql.validate(($) => $.age > 18)
// → { valid: true, errors: [] }The live playground is the best place to see dozens of other JSMQL examples.
The arrow function is never executed — jsmql() calls Function.prototype.toString() on it, strips the parameter list, and parses the body. That single trick gives you:
- Formatting for free. Prettier, oxfmt, and every other JS formatter indent and line-break your query like any other JavaScript. No jsmql plugin, no custom config.
- Linting for free. ESLint, Biome, and your editor's TypeScript service see real JS — they flag typos, unused identifiers, and shape mismatches at write time.
- Code completion. With
import "@koresar/jsmql/ops", your IDE autocompletes every stage and operator name, suggests the argument keys from the official MongoDB MQL spec, and surfaces the operator's description on hover. - AI coding works out of the box. Copilot, Cursor, and Claude already know JavaScript — they autocomplete jsmql idiomatically because jsmql is JavaScript. There is no new vocabulary for them to learn.
- Pre-compilation. jsmql.compile() parses once, executes many times.
- JS you already know — operators, ternaries, template literals, optional chaining, spread, computed keys, numeric separators,
Math.*,Date,typeof,instanceof, comments. Ifnode --checkaccepts it, jsmql does too. - 182 operators, full coverage — every aggregation expression and accumulator from the official MongoDB MQL spec, including Bitwise and Window categories. Unknown operators pass through, so new MongoDB releases work day one.
- Plain MQL passes through. Drop hand-written MQL JSON inline —
{ $gt: ["$age", 18] }, a whole stage, a whole pipeline — and jsmql compiles it to itself. Mix the two freely, migrate one expression at a time, or paste verbatim from the MongoDB docs. - Filter vs Pipeline picked automatically — a stage call (
$match(...),$project(...), …) or an update op ($.x = …) at the top level lowers as aPipeline(db.coll.aggregate(pipeline)/db.coll.updateOne(filter, update)); any;-separated input lowers as a multi-stage Pipeline; everything else lowers as aFilter(db.coll.find(filter)). Index-safe predicates translate to query-document form so existing indexes still get used. - Joins as JS —
$$$.<coll>.find(pred)/$$$.<coll>.filter(pred)lower to$lookupstages..find()follows JS semantics — returns one doc or null;.filter()keeps the array. Chained.length,.reduce(fn, init), and member access compose inline. Block-body lambdas (o => { $match(...); $sort(...); $limit(N); }) become the full sub-pipeline body.$$$$.<db>.<coll>.find/filter(pred)covers cross-database joins (requires Atlas Data Federation). Indexes still get used when the predicate is a simple field-to-field equality. See docs/LANGUAGE.md → Cross-collection lookups. - Collection unions as
Array.push—$$.push({...}, ...$$$.<coll>.filter(pred), $$$.<other>.find(pred))lowers to$unionWithstages. The spread (...) rule is JS-faithful:.filterand bare collections are arrays so they must be spread;.findand inline objects are scalars so they must not.$$$$.<db>.<coll>works for cross-database union (same Atlas caveat as cross-DB lookups). See docs/LANGUAGE.md → Collection union. - Replace root as JS assignment —
$ = <expr>lowers to$replaceWith: lift a sub-document ($ = $.profile), merge fresh fields ($ = { ...$, score: ... }— bare$is the current document, like MQL's$$ROOT), or pivot to a joined doc ($ = $$$.users.find(pred)). If the RHS clearly isn't a document (array literal, scalar,.filter()lookup), you get a compile-time error pointing at the fix. See docs/LANGUAGE.md → Replace root via$ = <expr>. - Materialised views via
$out—$$$.<coll> = $$and$$$$.<db>.<coll> = $$lower to a$outstage: the LHS names the destination, the RHS names the (optionally filtered) source. An inline$$.filter(<predicate>)on the RHS emits a$matchbefore the write ($$$$.dw.archive = $$.filter(u => !u.active)→[{ $match: … }, { $out: { db: "dw", coll: "archive" } }]). Bracket form ($$$["my-coll.v2"] = $$) addresses collection names that aren't valid JS identifiers. The compiler enforces "$outmust be last" at compile time. See docs/LANGUAGE.md →$out. $facetas a named object of filters — when every value of$ = { … }is a$$.filter(<lambda>), the same surface lowers to one$facetstage with each entry as a named sub-pipeline. Inside each lambda, the param is the input document (useo.<field>, not$.<field>); use an expression body for a simple predicate or a block body to run a full pipeline of stages. See docs/LANGUAGE.md → $facet via$ = { key: $$.filter(p), … }.- JS array mutators mutate at statement position —
$.events.sort(e => e.t),$.events.push(x),$.events.pop(),.shift(),.unshift(...),.reverse(),.splice(...),.fill(...)desugar to a$setstage that reassigns the field..toSorted(),.toReversed(),.toSpliced(...),.with(...)keep returning a new array. Sort key functions (.toSorted(e => e.distance),.toSorted(e => -e.distance)) lower to MongoDB'ssortBy: { path: ±1 }shape — including nested paths. Mutators in expression position throw with both the immutable variant and the statement-position option called out. See docs/LANGUAGE.md → Mutators. - Three call shapes — arrow
jsmql(($) => …), stringjsmql("…"), and template tagjsmql`…${val}…`for embedding outer-scope values. - Polymorphic by default, strict on demand —
jsmql()picks Filter or Pipeline from the input;jsmql.filter(),jsmql.pipeline(), andjsmql.update()lock it to one shape and throw an actionable error otherwise (with the offending stage named, forupdate()).jsmql.compile(fn)parses once for parameterised parse-once-bind-many.jsmql.expr()returns the raw aggregation expression that drops into a stage body. The three call shapes (string / arrow / template tag) apply to all of them. @koresar/jsmql/ops— a pure-types side-effect import that adds ambient$match/$dateAdd/ … globals. Zero runtime cost; bundlers tree-shake it to nothing.- Actionable errors — every error names the construct, suggests the nearest valid name (
Did you mean '…'?), and carries a real.posso editors can underline the offending region. - Strict TS, strippable source — runs as-is on Node 22.18+ / 24.3+, Deno, and Bun (no flags, no transpile).
A one-shot registration patches the Model static methods so the standard find / updateOne / aggregate / … calls accept jsmql source directly, alongside the plain MQL-JSON forms you already pass them:
const mongoose = require("mongoose");
require("@koresar/jsmql/mongoose")(mongoose);
// or, ESM: import jsmqlMongoose from "@koresar/jsmql/mongoose"; jsmqlMongoose(mongoose);
const User = mongoose.model("User", new mongoose.Schema({ name: String, age: Number, score: Number }));
User.find("$.age > 18"); // → find({ age: { $gt: 18 } })
User.find(($) => $.age > 18 && $.region === "AU"); // → find({ age: { $gt: 18 }, region: "AU" })
User.updateMany({}, ($) => $.score += 1);
// → updateMany({}, [{ $set: { score: { $add: ["$score", 1] } } }])
User.aggregate(($) => {
$match($.status === "active");
$group({ _id: $.region, total: { $sum: $.amount } });
$sort({ total: -1 });
});
User.find({ age: { $gt: 18 } }); // plain MQL JSON still passes through untouchedDetection rule. A patched argument is treated as jsmql source only when it's a string or a function. Plain objects/arrays (the regular MQL JSON forms) pass through to mongoose unchanged, so existing call sites need no migration. Template-tag inputs (jsmql\…``) lower to an object at the user's call site, so they take the pass-through path too.
TypeScript. The plugin ships a declare module "mongoose" augmentation that adds JSMQL-shaped overloads (string | JsmqlFn) to every patched Model static, so User.find("$.age > 18") and User.aggregate(($) => …) type-check after import "@koresar/jsmql/mongoose" — no per-call cast required. Mongoose's own FilterQuery<T> / UpdateQuery<T> overloads still apply on the MQL-JSON pass-through path.
Patched methods (with the slot used): find / findOne / findOneAnd{Delete,Replace,Update} / countDocuments / deleteOne / deleteMany / replaceOne / exists (filter at 0), updateOne / updateMany / findOneAndUpdate / findByIdAndUpdate (update at 1), distinct (filter at 1), aggregate (pipeline at 0). Each slot lowers through the matching strict-shape entry (jsmql.filter / jsmql.update / jsmql.pipeline), so a wrong-shape input — e.g. a bare expression at an aggregate slot — throws with the actionable strict-mode error at the patched call site instead of silently going wrong server-side. Registering twice on the same mongoose is a no-op.
See docs/specs/mongoose-plugin.md for the full per-slot table, the methods that are deliberately not patched (e.g. findOneAndReplace's replacement document), and the idempotence / subclass-propagation contracts.
- Live playground — write jsmql, see the MQL JSON update live. Pre-loaded with real-world recipes: tiered discounts, slug generation, audit logs, pivot tables, parameterised reports, and more.
- docs/LANGUAGE.md — the full language reference: every operator, every method, update-filter rules,
$matchquery translation,jsmql.compileparameter semantics,jsmql.exprfor raw aggregation expressions, the strict-shape entry points (jsmql.filter/jsmql.pipeline/jsmql.update), the@koresar/jsmql/opsimport, error catalogue, server-side-JS migration guide. - docs/DEVLOG.md — the running record of language decisions and the reasoning behind them.
MIT