The best feature flag tools in 2026
There's no single best feature flag tool; the right one depends on what you need most. LaunchDarkly is the broadest hosted platform, Unleash and Flagsmith are open source you can self-host, GrowthBook and Statsig lead on experiments, PostHog puts flags beside product analytics, ConfigCat is a focused hosted service with many SDKs, and Datadog and Cloudflare Flagship tie flags to their own platforms. Flagger.gg is a hosted flag service built for teams whose code AI agents write, with scoped agent tokens, approvals and a per-run audit trail.
Updated
Pick by what you need most
Every tool here can turn a feature on or off for chosen users without a deploy. They differ in what they're built around. Facts come from each vendor's own docs or repository, linked, checked on November 11, 2026. Pricing and plans are left out; check each vendor for what's in which plan. Flagger.gg here is the hosted flag service at flagger.gg, not the Flux/CNCF Flagger, a Kubernetes operator.
| If you need | Look at |
|---|---|
| The widest SDK coverage, experiments and metric-guarded rollouts in one hosted platform | LaunchDarkly |
| Open source you run on your own infrastructure | Unleash, Flagsmith, GrowthBook; or GO Feature Flag and FeatBit for something smaller |
| Experiments with real statistics | GrowthBook, Statsig, LaunchDarkly |
| Flags beside product analytics and session replay | PostHog, Statsig |
| A focused hosted flag service with SDKs for many languages, mobile and game engines | ConfigCat |
| Flags tied to the observability you already run | Datadog Feature Flags |
| Flags evaluated inside Cloudflare Workers | Cloudflare Flagship |
| Guardrails for AI coding agents that change flags: scoped tokens, approvals, a per-run audit trail | Flagger.gg (Go and Node) |
The tools, one by one
| Tool | What it's built around | Open source and hosting | Agent access |
|---|---|---|---|
| LaunchDarkly | Breadth: client-side, mobile, server-side, edge and AI SDKs (SDKs), experiments with multi-armed bandits (experimentation) and guarded rollouts that watch metrics and can roll back (guarded rollouts) | Hosted, with an open-source Relay Proxy you can run (Relay Proxy) | MCP server, hosted with OAuth or local (docs) |
| Unleash | Activation strategies, constraints and segments | Open source, AGPL-3.0 (repo); self-hosted, hybrid or Unleash Cloud (options) | MCP server, remote or local; files a change request where one is required (docs) |
| Flagsmith | Flags with remote config, segments and multivariate flags | Open source, BSD-3-Clause (repo); self-hosted, private cloud or SaaS | MCP server over Streamable HTTP with OAuth or an API key, or stdio (docs) |
| GrowthBook | Warehouse-native experiments, Bayesian or frequentist (overview) | Open core, MIT outside its enterprise directories (repo); GrowthBook Cloud, or self-hosted with Docker or Helm (self-host) | MCP server, hosted or self-run, with a Human-in-the-Loop default (docs) |
| PostHog | Flags next to product analytics, session replay and experiments | Open core, MIT outside ee/ (repo); PostHog Cloud, and its docs call self-hosted deployments unsupported (self-host) | Hosted MCP server that can work with feature flags (docs) |
| Statsig | Gates, dynamic configs, experiments and product analytics; server SDKs evaluate locally and poll every 10 seconds by default (server SDKs). Statsig's blog says Amplitude acquired it in May 2026 (Statsig) | Hosted | Hosted MCP server with 18 default tools, shown by your permissions (docs) |
| ConfigCat | A focused flag service with SDKs for many languages, including iOS, Android, Unity and Unreal Engine (SDKs); local evaluation, polling every 60 seconds by default (caching) | Open-source SDKs; the service is hosted, with an on-premise option (on-premise) | Local MCP server over stdio (docs) |
| Datadog Feature Flags | Flags on OpenFeature beside Datadog's telemetry, with canary rollouts watched by guardrail metrics (docs); from Datadog's 2025 purchase of Eppo (announcement) | Part of the Datadog platform | A feature-flags toolset in Datadog's MCP server (details) |
| Cloudflare Flagship | Flags evaluated inside Workers from config at the edge; in public beta since May 26, 2026 (changelog); a change can take up to 30 seconds to reach every location (concepts) | Part of Cloudflare | Wrangler commands with --json (details) |
| GO Feature Flag | Flags in a configuration file served by a stateless relay proxy, built on OpenFeature (docs) | Open source, MIT (repo); self-hosted | Not in the docs we checked |
| FeatBit | Flags, experimentation and remote config | Open source, MIT (repo); self-hosted | An official featbit-mcp repository (repo) |
| Flagger.gg | AI coding agents as first-class users: agent tokens scoped by project, environment and action; approvals with an agent rollout cap; a note and run on every agent write; Go and Node SDKs that evaluate in memory | Not open source; hosted only | MCP hosted at mcp.flagger.gg or flagger mcp on stdio, 11 tools shown by token scope and no approve tool; a CLI with --json on every command |
Closer looks: LaunchDarkly, Unleash, Flagsmith and PostHog, GrowthBook, Statsig and ConfigCat, Datadog and Cloudflare Flagship.
Questions that decide it
- Must you self-host? Then it's open source: Unleash, Flagsmith, GrowthBook, GO Feature Flag or FeatBit. Flagger.gg, Statsig and the platform-tied tools are hosted.
- Do you need experiments? GrowthBook, Statsig or LaunchDarkly. Experiments on Flagger.gg are In development.
- Which languages and platforms? For browsers, mobile or game engines, pick a tool with those SDKs: LaunchDarkly, ConfigCat, GrowthBook and others cover far more than Flagger.gg, which has Go and Node today; browser/React, Python and mobile SDKs are In development.
- Where does the check run? Local evaluation keeps the network out of every check. Most server SDKs here evaluate locally; what differs is how fast changes arrive. See local evaluation.
- Who changes your flags? If AI agents write and ship most of your code, look at what stops an agent from going too wide: how its token is scoped, whether a person approves wide changes, and whether each change records which agent and run made it.
When AI agents ship your code
Most tools above now publish an MCP server, and several add guardrails of their own: Unleash's files change requests, GrowthBook's defaults to Human-in-the-Loop and Statsig's shows tools by permission. Flagger.gg is built around the agent from the start:
- Agent tokens limited to one project, a list of environments and the actions
read create target rollout kill request, expiring after 90 days (scoped agent tokens). - An approvals mode in which agents roll out up to a cap in production, 10% by default, and a person approves anything wider; there's no approve tool for agents (approvals).
- A note on every agent write and the run recorded in the audit log, filterable by run (audit log).
- Protected flags that agents can only kill, and a kill that is exempt from agent rate limits.
It's built from the flag system that runs Sandboxer and Transformer in production, from North Bay Ventures.
FAQ
What are the best feature flag tools in 2026?
The ones teams compare most are LaunchDarkly, Unleash, Flagsmith, GrowthBook, PostHog, Statsig, ConfigCat, Datadog Feature Flags and Cloudflare Flagship, with GO Feature Flag and FeatBit as smaller open-source options. Flagger.gg is the one built for teams whose code AI agents write.
Which feature flag tools are open source?
Unleash (AGPL-3.0), Flagsmith (BSD-3-Clause), GO Feature Flag and FeatBit (both MIT) are open source, and GrowthBook and PostHog are open core. ConfigCat's SDKs are open source but its service isn't. Flagger.gg is not open source.
What's the best feature flag tool for a startup?
One you can set up in an afternoon that covers your stack. Self-host Unleash or Flagsmith if you must run it yourself; use PostHog if you already use it for analytics; use Flagger.gg if AI agents ship your Go or Node code. More in feature flags for startups.
Which feature flag tools have an MCP server?
LaunchDarkly, Unleash, Flagsmith, GrowthBook, PostHog, Statsig, ConfigCat and Datadog all document one, and FeatBit has an official repository. Flagger.gg's has 11 flag tools, shown only when the agent token's actions allow them.
Do I need a feature flag tool, or can I build my own?
A database table and an if statement work for a few flags. Teams usually move to a service when they need targeting, an audit log, local evaluation and safe access for many people and agents. See moving off an in-house flag system.
Ship your next change behind a flag.
Sign up, add an SDK key to your app and give your agent a token.