query

Neo4j Query API Go SDK

Overview

A Go client for the Neo4j Query API. Execute Cypher queries against Neo4j databases using plain HTTP — no Bolt protocol, no driver, no binary dependencies. The SDK mirrors the transformer pattern from the official Neo4j Go driver, so switching between them is straightforward.

result, err := query.WithTransformer(client.Query, ctx,
    "MATCH (n:Person) RETURN n.name AS name", nil,
    query.EagerResultTransformer)

Table of Contents

• Installation
• Quick Start
• Authentication
• Configuration
• API Flavors
• Access Mode
• Streaming
• Context and Timeouts
• Executing Queries
• Transformers
  • EagerResultTransformer
  • Collect
  • Single
  • Custom transformers
• Working with Records
• Error Handling
• Best Practices
• CI & Releases

---

Installation

go get github.com/neo4j-contrib/query-go-sdk

Quick Start

package main

import (
    "context"
    "fmt"
    "log"
    "time"

    query "github.com/neo4j-contrib/query-go-sdk"
)

func main() {
    client, err := query.NewClient(
        query.WithBasicAuth("neo4j", "password"),
        query.WithBaseURL("http://localhost:7474"),
        query.WithTimeout(30*time.Second),
    )
    if err != nil {
        log.Fatalf("Failed to create client: %v", err)
    }
    defer client.Close()

    ctx := context.Background()

    result, err := query.WithTransformer(
        client.Query,
        ctx,
        "MATCH (n:Person) RETURN n.name AS name LIMIT 10",
        nil,
        query.EagerResultTransformer,
    )
    if err != nil {
        log.Fatalf("Query failed: %v", err)
    }

    for _, rec := range result.Records {
        name, _ := rec.GetString("name")
        fmt.Printf("Name: %s\n", name)
    }
}

---

Authentication

Authentication is required and mutually exclusive — pass either WithBasicAuth or WithBearerToken, not both.

Basic authentication

client, err := query.NewClient(
    query.WithBasicAuth("neo4j", "your-password"),
)

Bearer token

client, err := query.NewClient(
    query.WithBearerToken("your-bearer-token"),
)

---

Configuration

Option	Default	Description
WithBaseURL(url)	http://localhost:7474	Base URL of the Neo4j server
WithDatabase(db)	—	Target database name
WithTimeout(d)	120s	Per-request timeout
WithMaxRetry(n)	3	Maximum retry attempts on transient failure
WithLogger(l)	warn to stderr	Structured *slog.Logger
WithHTTPClient(c)	default transport	Custom *http.Client (for mTLS, proxies, etc.)
WithUserAgent(ua)	query-go-sdk/<version>	Value of the User-Agent header
WithMaxResponseSize(n)	10MB	Maximum response body size in bytes; returns an error if exceeded
WithDefaultHeaders(map)	—	Extra headers sent with every request; Authorization, Content-Type, Accept, and User-Agent are protected and cannot be overridden
WithAPIFlavor(f)	FlavorQueryV2	Select which HTTP API endpoint to target; see API Flavors
WithAccessMode(m)	AccessModeUnset	Set the access mode sent with every request; see Access Mode
WithStreamingSupport(b)	false	Enable the Query API's streaming (JSON Lines) response format and unlock ExecuteStream; see Streaming

client, err := query.NewClient(
    query.WithBasicAuth("neo4j", "password"),
    query.WithBaseURL("http://neo4j.example.com:7474"),
    query.WithDatabase("mydb"),
    query.WithTimeout(60*time.Second),
    query.WithMaxRetry(5),
    query.WithDefaultHeaders(map[string]string{
        "X-Request-Source": "my-service",
    }),
)

---

API Flavors

The SDK supports two Neo4j HTTP APIs, selectable via WithAPIFlavor:

Constant	Endpoint	Response format	Default
query.FlavorQueryV2	/db/{db}/query/v2	Typed JSON ($type envelopes)	yes
query.FlavorLegacyHTTP	/db/{db}/tx/commit	Plain JSON row format	—

FlavorQueryV2 is the default and requires no explicit configuration.

Targeting the legacy HTTP API

Use FlavorLegacyHTTP to send queries to the older Cypher HTTP Transaction API. This is useful for performance comparison testing or for connecting to Neo4j versions that pre-date the Query API.

client, err := query.NewClient(
    query.WithBasicAuth("neo4j", "password"),
    query.WithBaseURL("http://localhost:7474"),
    query.WithAPIFlavor(query.FlavorLegacyHTTP),
)

Everything else — transformers, records, error handling — works identically for both flavors.

Type mapping differences

The legacy API returns row values as plain JSON without typed envelopes. Scalars (string, int64, float64, bool, nil) decode identically to FlavorQueryV2. However, graph entities are returned as plain map[string]any instead of *query.Node / *query.Relationship, because the /tx/commit endpoint does not include element IDs or labels in the row format. rec.GetNode and rec.GetRelationship will return (nil, false) for those values. The /tx/commit endpoint also doesn't return queryType/timing metadata, so Response.QueryType, ResultAvailableAfter, and ResultConsumedAfter stay at their zero values for FlavorLegacyHTTP.

---

Access Mode

The WithAccessMode option controls the access mode sent with every request. This hints to the server whether the operation requires write access or can be served by a read replica. With Neo4j Clusters, this option can be used to redirect queries to followers and mutations to the leader as the latter is the only cluster member who can make changes to the graph. This allows for the application to make load balancing decisions to avoid overwhelming the leader.

Constant
AccessModeUnset (default)
AccessModeRead
AccessModeWrite

By default, no access mode is sent at all. In a cluster, this means server-side routing decides where to send the request: any cluster member can handle it if it turns out to be a read, but only the leader can handle a write, so server-side routing forwards the request to the leader whenever the operation turns out to require a write.

// Read-only workload — may be routed to a read replica
client, err := query.NewClient(
    query.WithBasicAuth("neo4j", "password"),
    query.WithBaseURL("http://localhost:7474"),
    query.WithAccessMode(query.AccessModeRead),
)

// Write workload — explicitly forces leader routing
client, err := query.NewClient(
    query.WithBasicAuth("neo4j", "password"),
    query.WithBaseURL("http://localhost:7474"),
    query.WithAccessMode(query.AccessModeWrite),
)

---

Streaming

By default, Execute buffers the entire query result before returning it. WithStreamingSupport(true) switches the wire format to the Query API's streaming (JSON Lines) response and unlocks QueryService.ExecuteStream, which decodes records incrementally as they arrive instead of holding the full result in memory. This matters for large result sets, and lets a consumer start processing rows before the server has finished sending them.

client, err := query.NewClient(
    query.WithBasicAuth("neo4j", "password"),
    query.WithBaseURL("http://localhost:7474"),
    query.WithStreamingSupport(true),
)

ExecuteStream returns a *query.StreamResult. Iterate over Records() with a standard Go range loop; the same *query.Record accessors described in Working with Records apply to each row:

result, err := client.Query.ExecuteStream(ctx, "MATCH (n:Person) RETURN n.name AS name", nil)
if err != nil {
    log.Fatal(err)
}

fmt.Println(result.Fields()) // ["name"] — available immediately, before any rows

for rec, err := range result.Records() {
    if err != nil {
        log.Fatal(err) // transport/decode error, or a *query.QueryErrors surfaced mid-stream
    }
    name, _ := rec.GetString("name")
    fmt.Println(name)
}

summary := result.Summary() // bookmarks, notifications, timings — populated once Records() is fully drained
fmt.Println(summary.Bookmarks)

StreamSummary carries the same QueryType, ResultAvailableAfter, and ResultConsumedAfter fields as query.Response/EagerResult — Neo4j reports this metadata identically whether or not streaming is enabled.

Breaking out of the loop early still releases the underlying HTTP connection — Records() closes it automatically whether iteration ends naturally, on error, or via an early break. Call result.Close() directly if you need to abandon a stream without iterating at all.

WithStreamingSupport is a client-wide setting: Execute is unaffected by it, and ExecuteStream returns an error if the client wasn't constructed with it enabled. It is not supported together with WithAPIFlavor(query.FlavorLegacyHTTP) — NewClient returns an error if both are set, since the legacy Cypher HTTP Transaction API has no streaming format.

Because ExecuteStream returns before the response body has been fully read, the request's timeout (WithTimeout, default 120s) bounds the entire stream lifetime — from the initial request through however long the caller takes to finish draining Records() — not just the time to the first byte. Raise WithTimeout for clients that stream very large results or consume them slowly.

---

Context and Timeouts

Every operation accepts a context.Context as its first argument. The service applies its own timeout via context.WithTimeout on every call — whichever deadline is shorter (parent or service) wins.

// Per-call deadline overrides the service timeout when shorter
ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
defer cancel()

resp, err := client.Query.Execute(ctx, "MATCH (n) RETURN count(n)", nil)

// Cancellation propagates through to in-flight HTTP requests
ctx, cancel := context.WithCancel(context.Background())
go func() { <-shutdown; cancel() }()

_, err := client.Query.Execute(ctx, "MATCH (n) RETURN n", nil)
if errors.Is(err, context.Canceled) {
    log.Println("query cancelled")
}

---

Executing Queries

Direct execution

client.Query.Execute returns a *query.Response containing the raw decoded result.

resp, err := client.Query.Execute(ctx, "RETURN 1 AS n", nil)
if err != nil {
    log.Fatal(err)
}

fmt.Println(resp.Fields)    // ["n"]
fmt.Println(resp.Rows)      // [[int64(1)]]
fmt.Println(resp.Bookmarks)

query.Response fields:

Field	Type	Description
Fields	[]string	Ordered column names
Rows	[][]any	Decoded row values (see type mapping below)
Notifications	[]query.Notification	Advisory messages from Neo4j
Bookmarks	[]string	Transaction bookmarks
QueryPlan	*query.PlanOperator	Non-nil for EXPLAIN/PROFILE queries
QueryType	string	"r" (read), "rw" (read/write), "w" (write), or "s" (schema write)
ResultAvailableAfter	time.Duration	Time for the result to become available
ResultConsumedAfter	time.Duration	Time to fully consume the result

Passing parameters

params := map[string]any{"name": "Alice", "age": 30}
resp, err := client.Query.Execute(ctx,
    "MATCH (n:Person {name: $name}) WHERE n.age >= $age RETURN n.name",
    params)

---

Transformers

The transformer pattern mirrors the Neo4j Go driver's neo4j.ExecuteQuery. A transformer is a function that converts *query.Response into a typed Go value:

type ResultTransformer[T any] func(*query.Response) (T, error)

Three built-in transformers cover the common cases. You can also write your own — see Custom transformers.

result, err := query.WithTransformer(svc, ctx, cypher, params, transformer)

EagerResultTransformer

Collects all rows into a *query.EagerResult. Use for typical queries where the full result set fits in memory.

result, err := query.WithTransformer(
    client.Query, ctx,
    "MATCH (n:Person) RETURN n.name AS name, n.age AS age",
    nil,
    query.EagerResultTransformer,
)
if err != nil {
    log.Fatal(err)
}

fmt.Println(result.Keys)          // ["name", "age"]
fmt.Println(len(result.Records))  // number of rows

for _, rec := range result.Records {
    name, _ := rec.GetString("name")
    age, _  := rec.GetInt64("age")
    fmt.Printf("%s is %d\n", name, age)
}

EagerResult also exposes HasWarnings() and Warnings() for advisory notifications, and the same QueryType, ResultAvailableAfter, and ResultConsumedAfter fields documented above for query.Response.

Collect

Maps each row to a value using a function, returning []T.

type Person struct {
    Name string
    Age  int64
}

people, err := query.WithTransformer(
    client.Query, ctx,
    "MATCH (n:Person) RETURN n.name AS name, n.age AS age",
    nil,
    query.Collect(func(rec *query.Record) (Person, error) {
        name, _ := rec.GetString("name")
        age, _  := rec.GetInt64("age")
        return Person{Name: name, Age: age}, nil
    }),
)

Single

Expects exactly one row; returns an error if zero or more than one row is returned.

type Movie struct {
    Title    string
    Released int64
}

movie, err := query.WithTransformer(
    client.Query, ctx,
    "MATCH (m:Movie {title: $title}) RETURN m.title AS title, m.released AS released",
    map[string]any{"title": "The Matrix"},
    query.Single(func(rec *query.Record) (Movie, error) {
        title, _    := rec.GetString("title")
        released, _ := rec.GetInt64("released")
        return Movie{Title: title, Released: released}, nil
    }),
)

Custom transformers

Because *query.Response is part of the public API, you can write your own ResultTransformer for cases that EagerResultTransformer, Collect, and Single don't cover.

// A transformer that extracts a single column as a flat []string
func firstColumnStrings(resp *query.Response) ([]string, error) {
    out := make([]string, 0, len(resp.Rows))
    for _, row := range resp.Rows {
        if len(row) == 0 {
            continue
        }
        s, ok := row[0].(string)
        if !ok {
            return nil, fmt.Errorf("expected string, got %T", row[0])
        }
        out = append(out, s)
    }
    return out, nil
}

names, err := query.WithTransformer(
    client.Query, ctx,
    "MATCH (n:Person) RETURN n.name",
    nil,
    query.ResultTransformer[[]string](firstColumnStrings),
)

Custom transformers work with any named function whose signature matches func(*query.Response) (T, error).

---

Working with Records

*query.Record provides named field access that mirrors the Neo4j Go driver's neo4j.Record.

Typed accessors

s, ok    := rec.GetString("name")          // string
n, ok    := rec.GetInt64("count")          // int64
f, ok    := rec.GetFloat64("score")        // float64
b, ok    := rec.GetBool("active")          // bool
bs, ok   := rec.GetBytes("data")           // []byte
t, ok    := rec.GetTime("createdAt")       // time.Time
d, ok    := rec.GetDuration("lifespan")    // query.Duration
p, ok    := rec.GetPoint("location")       // query.Point
node, ok := rec.GetNode("n")               // *query.Node
rel, ok  := rec.GetRelationship("r")       // *query.Relationship
path, ok := rec.GetPath("p")               // query.Path
list, ok := rec.GetList("tags")            // []any
m, ok    := rec.GetMap("props")            // map[string]any

All accessors return (zero, false) when the field is absent or null.

Graph entity types

query.Node, query.Relationship, and query.Path are part of the public API — you can use them in your own function signatures:

func nodeLabel(n *query.Node) string {
    if len(n.Labels) > 0 {
        return n.Labels[0]
    }
    return ""
}

func relSummary(r *query.Relationship) string {
    return fmt.Sprintf("[%s]", r.Type)
}

query.Node fields: ElementID string, Labels []string, Properties map[string]any.
query.Relationship fields: ElementID, StartNodeElementID, EndNodeElementID string, Type string, Properties map[string]any.

Neo4j → Go type mapping

Neo4j type	Go type
Null	nil
Boolean	bool
Integer	int64
Float	float64
String	string
ByteArray	[]byte
List	[]any
Map	map[string]any
Date / Time / DateTime / LocalTime / LocalDateTime	time.Time
Duration	query.Duration
Point (2D / 3D)	query.Point
Node	*query.Node
Relationship	*query.Relationship
Path	query.Path

Generic accessor

val, ok := rec.Get("fieldName")  // returns (any, bool)

Other helpers

rec.Keys()           // []string — ordered field names
rec.Values()         // []any   — ordered decoded values
rec.AsMap()          // map[string]any
rec.IsNull("field")  // true if field present and null

List helpers

rawList, _ := rec.GetList("tags")
tags, ok   := query.StringList(rawList)   // []string
counts, ok := query.Int64List(rawList)    // []int64
scores, ok := query.Float64List(rawList)  // []float64

---

Error Handling

HTTP errors — *query.Error

Returned when the Neo4j server responds with a non-2xx status code.

resp, err := client.Query.Execute(ctx, cypher, params)
if err != nil {
    var apiErr *query.Error
    if errors.As(err, &apiErr) {
        fmt.Printf("HTTP %d: %s\n", apiErr.StatusCode, apiErr.Message)

        switch {
        case apiErr.IsUnauthorized():
            log.Fatal("check credentials")
        case apiErr.IsNotFound():
            log.Fatal("endpoint not found")
        case apiErr.IsBadRequest():
            log.Fatal("bad request")
        }
    }
}

query.IsNotFound(err) is a convenience helper that unwraps the error chain.

Neo4j query errors — *query.QueryErrors

Returned when Neo4j executes the request but reports one or more Cypher errors (syntax errors, constraint violations, etc.).

resp, err := client.Query.Execute(ctx, "INVALID CYPHER", nil)
if err != nil {
    var qErr *query.QueryErrors
    if errors.As(err, &qErr) {
        for _, e := range qErr.Errors {
            fmt.Printf("[%s] %s: %s\n", e.Classification(), e.Title(), e.Message)
        }
    }
}

QueryError helper methods: Classification() (e.g. ClientError), Category() (e.g. Statement), Title() (e.g. SyntaxError).

Context errors

ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
defer cancel()

_, err := client.Query.Execute(ctx, cypher, nil)
if errors.Is(err, context.DeadlineExceeded) {
    log.Println("query timed out")
}
if errors.Is(err, context.Canceled) {
    log.Println("query cancelled")
}

---

Best Practices

Secure credentials

user := os.Getenv("NEO4J_USERNAME")
pass := os.Getenv("NEO4J_PASSWORD")

client, err := query.NewClient(
    query.WithBasicAuth(user, pass),
    query.WithBaseURL(os.Getenv("NEO4J_URL")),
)

Context management

Always pass a context with an appropriate deadline or cancellation. Use defer client.Close() to release idle connections when the client is no longer needed.

Error classification

Use errors.As to branch on *query.Error (HTTP-level) versus *query.QueryErrors (Cypher-level). Transient server errors (5xx) may be worth retrying; client errors (4xx) generally should not be.

---

CI & Releases

Three GitHub Actions workflows manage CI and the release process.

Workflows

Workflow	Trigger	What it does
CI	Push to main, every PR	Runs tests with the race detector, golangci-lint, and go build ./...
Changelog check	Every PR	Fails if the PR changes .go files but has no entry in .changes/unreleased/
Release	Push of a vX.Y.Z tag	Gates on tests, extracts the changelog section, creates a GitHub Release

Making a release

Releases follow a three-step process. changie collects the unreleased fragment files and determines the correct semver bump automatically from the change kinds (Added → minor, Fixed/Security → patch, Changed/Removed → major).

There is no manual version bump required. ClientVersion uses debug.ReadBuildInfo() at runtime, scanning the consumer's dependency list (info.Deps) for this module's own entry to read the version the Go toolchain embedded when the consumer built their application — info.Main describes the consumer's own application, not this SDK, so it's deliberately not used. It falls back to "development" only when this module can't be found in Deps at all, e.g. local and test builds of this repository itself.

1. Batch and merge the changelog

changie batch   # collects .changes/unreleased/*.yaml → .changes/vX.Y.Z.md
changie merge   # folds that file into CHANGELOG.md

2. Commit and tag

git add CHANGELOG.md .changes/
git commit -m "chore: release vX.Y.Z"
git tag vX.Y.Z
git push origin main --tags

3. Workflow takes over

Pushing the tag fires the Release workflow, which:

• Runs go test -race ./... — the release is aborted if any test fails
• Extracts the ## vX.Y.Z section from CHANGELOG.md
• Creates a GitHub Release with that text as the release notes

Because this is a Go module with no compiled binaries, the tag itself is what consumers reference:

go get github.com/neo4j-contrib/query-go-sdk@vX.Y.Z

Adding a changelog entry

Every PR that changes Go source files needs a changie fragment. Run:

changie new

Choose a kind and write a one-line summary, then commit the generated .yaml file alongside your code changes. The Changelog check workflow will fail the PR if this step is skipped.

To bypass the check for a PR that genuinely needs no entry (docs-only, CI-only, or test-only changes), add the no-changelog label to the PR.

---

License

See LICENSE file for details.