Full-text search in MongoDB

This chapter follows Transactions in MongoDB.

Regex and $text on self-hosted Community

Regex on arbitrary fields ({ title: { $regex: /mongo/i } }) is flexible but often cannot use indexes efficiently unless anchored to a prefix.

Text indexes support the $text operator and a single text index per collection:

1
db.articles.createIndex({ title: 'text', body: 'text' })
2
db.articles.find({ $text: { $search: 'aggregation framework' } })

Text search is language-aware but limited compared to dedicated search engines—no rich facets, fuzzy tuning, or autocomplete in one package.

Use regex and $text for simple filters; move to Atlas Search when relevance ranking and analyzers matter.

Atlas Search overview

Atlas Search (Lucene-backed) runs on Atlas and on Atlas Local (mongodb/mongodb-atlas-local) for development. It uses search indexes separate from standard MongoDB indexes and the $search aggregation stage.

Run Atlas Local for this chapter:

1
docker run -p 27017:27017 -p 8080:8080 mongodb/mongodb-atlas-local

Connect with the URI printed in the container logs.

Define a search index

Example index JSON (stored on the search node):

1
{
2
  "mappings": {
3
    "dynamic": false,
4
    "fields": {
5
      "title": { "type": "string", "analyzer": "lucene.standard" },
6
      "body": { "type": "string", "analyzer": "lucene.english" }
7
    }
8
  }
9
}

Create from TypeScript on Atlas (Search Index API or createSearchIndexes command depending on deployment):

1
await db.collection('articles').createSearchIndex({
2
  name: 'articles_default',
3
  definition: {
4
    mappings: {
5
      dynamic: false,
6
      fields: {
7
        title: { type: 'string', analyzer: 'lucene.standard' },
8
        body: { type: 'string', analyzer: 'lucene.english' },
9
      },
10
    },
11
  },
12
});

Wait until index status is READY before querying.

$search queries

Basic text search:

1
db.articles.aggregate([
2
  {
3
    $search: {
4
      index: 'articles_default',
5
      text: { query: 'vector embeddings', path: ['title', 'body'] },
6
    },
7
  },
8
  { $limit: 10 },
9
  { $project: { title: 1, score: { $meta: 'searchScore' } } },
10
])

Compound boolean queries combine must/should/mustNot:

1
{
2
  $search: {
3
    index: 'articles_default',
4
    compound: {
5
      must: [{ text: { query: 'mongodb', path: 'title' } }],
6
      should: [{ text: { query: 'atlas', path: 'body', fuzzy: { maxEdits: 1 } } }],
7
    },
8
  },
9
}

Autocomplete uses the autocomplete field type in the index definition and matching operator in queries.

Add highlights:

1
{
2
  $search: {
3
    index: 'articles_default',
4
    text: { query: 'pagination', path: 'body' },
5
    highlight: { path: 'body' },
6
  },
7
}

Faceted counts with $searchMeta:

1
{
2
  $searchMeta: {
3
    index: 'articles_default',
4
    facet: {
5
      operator: { text: { query: 'database', path: 'body' } },
6
      facets: { categoryFacet: { type: 'string', path: 'category', numBuckets: 10 } },
7
    },
8
  },
9
}

Project searchScore via { $meta: 'searchScore' } to sort by relevance.

Paginate with searchAfter

For stable cursor pagination through search results, use searchAfter with the sort values from the previous page (similar in spirit to keyset pagination):

1
{
2
  $search: {
3
    index: 'articles_default',
4
    text: { query: 'mongodb', path: 'title' },
5
    sort: { score: { $meta: 'searchScore' }, _id: 1 },
6
    searchAfter: '<token from previous page>',
7
  },
8
}

Next: Vector search and embeddings in MongoDB.

Full-text search in MongoDB

Regex and $text on self-hosted Community

Atlas Search overview

Define a search index

$search queries

Highlights, facets, and scoring

Paginate with searchAfter

Check your understanding

Search articles