Quick Start - AI coding agent
The fastest way to add Featureflow to a project is to hand the job to your AI coding agent — Claude Code, Cursor, Copilot, Codex or any other tool that reads agent skills.
Featureflow skills are instructions, not code. Once installed, your agent loads the relevant one on its own when you ask for something it covers: installing an SDK, putting code behind a flag, rolling it out, measuring it, and deleting it afterwards. Each skill carries the parts that are easy to get wrong — which key is secret and which is public, why the client is a singleton, why bucketing is stable, what order rules match in.
1. Install the skills
- Any agent (skills CLI)
- Claude Code plugin
- Manual
Run this in the root of your project:
npx skills add featureflow/featureflow-skills -y
This copies every Featureflow skill into your project. To install just one, name it:
npx skills add featureflow/featureflow-skills --skill featureflow-setup -y
Inside Claude Code:
/plugin marketplace add featureflow/featureflow-skills
/plugin install featureflow@featureflow
The plugin is available in every project, and updates come through /plugin.
Copy the skills into the project, or into ~/.claude/skills to have them everywhere:
git clone https://github.com/featureflow/featureflow-skills.git
cp -r featureflow-skills/skills/* .claude/skills/
Keep each skill's references/ directory next to its SKILL.md — the skills load those files
by relative path.
For tools without a skills directory, the skills are plain Markdown with YAML frontmatter. Point
your tool's rules or context mechanism at skills/*/SKILL.md, or paste the one you need into
the conversation.
2. Give the agent your SDK keys
Every environment has two SDK keys, found in the Featureflow dashboard under your project's environment → API Keys:
- Server SDK key (
sdk-srv-env-…) — secret. Server-side code only; keep it in an environment variable. - Client SDK key — public. Safe in browser and mobile code.
Paste them into your prompt, or tell the agent where to find them. The skills instruct it to keep keys out of anything you commit.
The first time you sign in, the dashboard shows a ready-made prompt with your keys already in it. Copy it into your agent and the dashboard confirms the moment your app connects.
3. Ask for the outcome
Ask for the outcome; the agent picks the skill.
Add Featureflow to this app
Detects the stack, installs the right SDK, wires a single client from an environment variable, and proves it with one real evaluation.
Put the new checkout behind a flag
Creates new-checkout off in every environment, wraps the smallest safe boundary, keeps the old
path intact, tests both branches, and tells you how to turn it on.
Roll it out to 10%
Explains that buckets are stable and ramps are additive, orders the rules so your excluded accounts stay excluded, and makes the change — or walks you through the dashboard for anything touching production.
Which flags can we delete?
Greps the codebase, reconciles against what Featureflow knows, and reports what is safe to remove — flagging the mobile clients and other repos that might still read a flag before anything is deleted.
The skills
| Skill | Use it when |
|---|---|
| featureflow-setup | Adding Featureflow to a project for the first time — pick the right SDK, wire the client, handle keys safely |
| featureflow-add-flag | Putting a new code path behind a flag, with both branches working and a removal plan |
| featureflow-targeting | Rolling out by percentage, targeting a segment, ordering rules, or reaching for the kill switch |
| featureflow-experiments | A/B testing a change and tracking conversions against variants |
| featureflow-cleanup-flags | Auditing which flags a codebase uses, and retiring the ones that are done |
| featureflow-mcp | Connecting the Featureflow MCP server so the agent can read and change real flag state |
Languages covered: Java · Node · browser JavaScript/TypeScript · React · React Native · Python · Ruby · Go · .NET. Per-SDK install, client construction, evaluation idioms and testing notes live in the featureflow-setup references.
Let the agent change flags too
The skills work on their own. With the Featureflow MCP server connected, your agent can also read and change real flag state — create the flag, turn it on in staging, ramp the rollout — instead of asking you to do it in the dashboard. Set it up from Administration → Integrations → AI assistants (MCP).
Prefer to do it yourself?
The step-by-step guides cover the same ground by hand:
Links
- featureflow-skills on GitHub — corrections welcome, especially where a skill describes an SDK inaccurately
- Featureflow MCP server