Unleash MCP Server: Setup for Claude, Cursor and Other Apps

• Finds flags across all projects by part of a name

  Looks up active flags in every project by part of their name or description. It returns the matches with their project, type and whether each environment has them on.

  Try: “Find the flags with checkout in the name.”

  Network · search_flags

• Lists the flags in one project

  Returns the flags in a project, newest or oldest first. It shows active flags by default and can list archived ones instead.

  Try: “List the archived flags in the payments project.”

  Network · list_flags

• Lists the Unleash projects your token can see

  Returns each project's ID, name, description and creation time. The assistant uses it to pick the right project before it touches a flag.

  Try: “List my Unleash projects.”

  Network · list_projects

• Shows a flag's settings and rollout rules

  Fetches one flag with its environments, rollout strategies and parent flags. It can also include what each segment targets and which context fields rules can use.

  Try: “Show me how new-checkout is set up in production.”

  Network · get_flag_state

• Shows who changed a flag and when

  Returns the recent events on a flag as a diff, with who made each change, when, and in which environment. Handy before you edit a flag or to explain why it is on.

  Try: “What changed on new-checkout this week?”

  Network · get_flag_history

• Checks whether a flag is on for a given user

  Runs a flag against a context such as a user ID in one environment. It reports the result and which strategies, constraints and segments matched. Unleash marks this API as alpha.

  Try: “Does user 42 get new-checkout in production?”

  Network · evaluate_flag

• Helps decide if a code change needs a flag

  Returns guidance on whether a change is risky enough for a feature flag, including how to spot existing parent flags. It reads nothing from Unleash and changes nothing.

  Try: “Should this checkout rewrite go behind a feature flag?”

  Read · evaluate_change

• Gives code for putting a change behind a flag

  Returns language-specific templates and rules for wrapping a change in a flag check. The check must sit inside the code path, not around route setup.

  Try: “Wrap this new export button in a feature flag.”

  Read · wrap_change

• Gives steps to remove a flag from your code

  Returns step-by-step instructions for deleting a flag's checks from the code while keeping the winning code path. It only guides the assistant. Your code changes through the assistant's own file tools.

  Try: “Remove the old-banner flag from the codebase.”

  Read · cleanup_flag

• Creates a new feature flag in Unleash

  Creates a flag in a project with a name, type and description. The assistant is told to follow Unleash's naming and lifecycle advice.

  Try: “Create a release flag called new-checkout in the shop project.”

  Write · create_flag

• Creates context fields and segments for rollouts

  Creates the context fields and segments that rollout rules point at, in one project or in every project. Each item is a separate call and nothing is rolled back if one fails partway.

  Try: “Create a beta-testers segment for user IDs 1, 2 and 3.”

  Write · create_targeting

• Sets who gets a flag in each environment

  Sets a rollout percentage, optional segments and context-field rules, and whether the environment is on. Where an environment needs approval, it opens a change request instead. The tool is flagged as destructive.

  Try: “Roll out new-checkout to 10% of users in production.”

  Write · set_flag_rollout

• Turns a flag on or off in one environment

  Enables or disables a flag in a named environment. Where changes need approval, it files a change request instead.

  Try: “Turn off new-checkout in production.”

  Write · toggle_flag_environment

• Deletes a rollout rule from a flag

  Deletes one strategy from a flag's environment. Where changes need approval, it files a change request and the strategy stays until that is approved.

  Try: “Remove the old 50% rollout rule from new-checkout in staging.”

  Delete · remove_flag_strategy

• Completes, archives or revives a flag

  Marks a flag complete, undoes that, archives it, revives it, or sets and clears the stale marker. Archiving stops SDKs from receiving the flag, and is refused if other flags depend on it.

  Try: “Archive the old-banner flag.”

  Delete · update_flag_lifecycle

• Changes a project's flag defaults

  Sets a project's default stickiness, adds environments to it, and sets each environment's default strategy. The default strategy also applies to existing flags that have none. There is no rollback.

  Try: “Add the staging environment to the shop project.”

  Write · update_project_settings

• Used by the assistant behind the scenes

  Guides a search for existing flags in your code

  Returns instructions for finding flags that already exist in the code, files and git history. The assistant uses it to avoid duplicates before it creates a new flag.

  Try: “Is there already a flag for the new search page?”

  Read · detect_flag

• Reports a failed or missing feature to Unleash

  Sends Unleash a report when a tool failed, a request was not supported or a result looked wrong. It asks you first. It is not listed if feedback is turned off in settings.

  Try: “(The assistant calls this on its own after asking you.)”

  Network · send_feedback

Watch out for

• It can delete rollout rules and archive flags

  One tool deletes a flag's strategy and another archives flags, which stops SDKs from getting them. Use a token with limited rights, or approve each call by hand. github.com

• One call can change what users see

  Rollout and toggle tools can switch a flag on for real users in production. Where an environment needs approval, they open a change request instead. github.com

• Project defaults apply to existing flags

  Setting an environment's default strategy also applies to flags that have none. Confirm values like 100% in production before the assistant sends them. github.com

• It can send feedback reports to Unleash

  A report names the tool, the error and your client. Nothing is sent until you agree, and you can turn the tool off with a setting. github.com

Security advisories