Introducing the DevCycle MCP Server: Ship Flags Faster, Safely with AI Coding Agents

Why we built a MCP server for DevCycle

AI coding environments and agents are great at writing code—but the moment you need to create a feature flag, tweak targeting, or check an audit log, you’re back to context‑switching: open the dashboard, copy keys, run a CLI, paste results, repeat. That’s slow, error‑prone, and kills flow.

The new DevCycle Model Context Protocol (MCP) Server lets AI‑powered editors (Cursor, VS Code, Windsurf) and assistants like Claude Code talk directly to your DevCycle projects to do real work for you: create flags, manage variations and targeting, fetch SDK keys, pull audit history, and even report on evaluation analytics. No tab‑sprawl. No manual stitching of disparate contexts. Just ask, and it acts safely and with guardrails.

What we built

A hosted MCP server for DevCycle. It’s based on our CLI and exposed as a secure, OAuth‑backed MCP endpoint. Connect once; your client negotiates streaming (SSE) or HTTP, and you’re ready to go.

• Hosted endpoints
  • Auto‑negotiating: https://mcp.devcycle.com/mcp
  • SSE‑only (fallback): https://mcp.devcycle.com/sse
• Broad tool coverage
  • Feature and variation CRUD, targeting rules, variables, environments, and SDK keys, project introspection, self‑targeting overrides, and usage/evaluation analytics.
• Production safety
  • Tools that can change production are clearly marked, and destructive actions require explicit confirmation.
  • The authentication ensures the MCP server respects the given user's permission level.

The problem DevCycle MCP solves (and how)

Before: Every time you create a new feature, you need to juggle whether to start in your IDE with the code or in the DevCycle interface to create the feature flag you're going to use. This is common with many actions and requires context switching. Sometimes, if you're using AI agents, you need to correct them when they do something before you're ready.

Now: your AI agent has direct access to DevCycle to take actions on your behalf, exactly when they make sense during the workflow.

You can ask things like:

• “Create a new feature flag called new-checkout-flow.”
• “Enable targeting for header-redesign in production.”
• “Show me evaluations for pricing-experiment from the last 7 days.”

Under the hood, your agent speaks to our server via the MCP, which executes the appropriate DevCycle commands with your authorization and returns structured results your agent can reason about.

Check out some full-fledged use cases here ⬇️

Why it’s unique

• OpenFeature‑native platform: DevCycle is the first OpenFeature‑native feature flagging platform—built by OpenFeature governance contributors—so your MCP workflows align with the standard you already use in code.
• Zero setup: It’s hosted and OAuth‑backed—no daemons, no ports. Connect and go; as long as you're authenticated, it's always running.
• Wide, practical surface area: From flag creation to audit trails and evaluations, the toolset maps to day‑to‑day release management—not just cool tech demos.
• Built for safety: Tied to your permission level via OAuth, with clear prod/destructive markers and confirmation flows. This ensures AI can't make destructive changes without your approval.

DevCycle MCP benefits at a glance

• Fewer context switches, faster cycles: Stay in your editor/assistant and speak the task; it handles the rest.
• Safer changes, clearer history: Toggle targeting or update rules with confirmations, then trace everything with audit logs.
• Better integration: Because the DevCycle MCP can run locally or remotely, you can easily create your own service that leverages the MCP for customized workflows, like cross-referencing an Observability tool MCP with the DevCycle MCP to confirm if a flag change caused an incident.
• Standards‑aligned: Plays nicely with OpenFeature SDKs and providers across your stack.

Get started (in under 2 minutes)

Pick your favorite AI agent and follow the instructions. For more detailed getting-started instructions, you can check out our docs.

Cursor

Or add to ~/.cursor/mcp_settings.json:

{
  "mcpServers": {
    "DevCycle": { "url": "https://mcp.devcycle.com/mcp" }
  }
}

Then open Cursor Settings → Rules and Integrations → Click Needs Login to authorize (select your org if prompted).

VS Code

Or add to .continue/config.json:

{
  "mcpServers": {
    "DevCycle": { "url": "https://mcp.devcycle.com/mcp" }
  }
}

Start the server in VS Code’s MCP panel, accept the auth dialog, and finish in your browser.

Claude Code

1. Open your terminal and run:

claude mcp add --transport http devcycle https://mcp.devcycle.com/mcp

2. In Claude Code, type:

/mcp

Select devcycle → Enter to login and complete the OAuth flow in your browser. When you return, the server will show as connected.

Prefer running the MCP locally? Install the CLI (npm i -g @devcycle/cli) and point clients at the bundled local dvc MCP server. Env‑var auth is supported for CI/CD.

What can you ask the DevCycle MCP to do?

The DevCycle MCP provides a complete selection of tools and prompts that together enable these popular use cases, among others:

Feature Creation

When creating a new feature in your code, you typically have to go back and forth between your IDE and DevCycle to create a Feature, get the relevant Variable for your code, implement that Variable and then wrap your code in the conditional. If you're using a coding agent to create the feature, it may make incorrect decisions about how to implement feature flags.

With the DevCycle MCP installed, your coding agent can handle the end-to-end of creating the Feature and Variable in DevCycle and then implement it in code to flag the new feature it started coding for you.

QA Testing

The DevCycle MCP enables natural language commands for configuring flags with DevCycle's self-targeting overrides. This gives your coding agent an understanding of the overrides set for your user and the ability to configure overrides on your behalf.

Testing a new feature locally is as easy as telling your coding agent to toggle a flag on or off, or you can give it more complex instructions on how to configure multiple flags in tandem for you, all without updating any targeting rules.

Audit Logging / Incident Research

Let's say an incident just kicked off, and you're in the middle of trying to determine if a code deploy or a feature flag configuration change may have caused the issue. Instead of hunting down relevant audit logs in DevCycle, you can now ask your coding assistant to list all changes to the production environment within a 30-minute range.

The DevCycle MCP will respond with all audit logs across all Features in your project filtered by your criteria, and then your coding agent can present that in an easy-to-digest way. Suppose you are running the MCP of an Observability tool in tandem. In that case, you can even have your coding agent cross-reference the errors to determine if any of the changes could plausibly have caused the issue, helping you and your team get to mitigation and resolution as quickly as possible.

Flag Cleanup

Cleaning up flags that are no longer needed is a tedious task that is often ignored in favor of new projects. This means that most teams that implement feature flags tend to keep those conditionals around longer than they would typically like. At DevCycle, we have already done a lot to help alleviate this problem, from providing Stale Feature Notifications to enabling easy automated cleanup of Variables via our CLI.

The DevCycle MCP is aware of the statuses of all of your DevCycle Features, so now you can have it take all Completed Features, read the values that are being distributed, and have your coding agent replace the stale Variables in your code with the relevant static values. All you have to do is review the code.

These are just a few examples of what you can do with the DevCycle MCP. To view all of the available tools, check out our reference docs.

Try it out now

Once you have installed and authenticated the DevCycle MCP, try asking your coding agent to:

• “Create a new feature flag called new-checkout-flow.”
• “List all features in my project.”
• “Enable targeting for header-redesign in development.”
• “Show evaluation analytics for the last 7 days.”
• "Show me all flag changes on August 18."

About DevCycle

DevCycle is the first OpenFeature‑native feature flag management platform - so you can ship faster without vendor lock‑in. The MCP server extends that developer‑first experience right into your AI tools.