Integration · Split.io / Harness FME

Feature flags via the Harness platform MCP.

Split.io is now part of Harness. AI access happens via the Harness platform MCP, scoped to the FME (Feature Management & Experimentation) toolset.

The Harness MCP exposes 137 resource types across the entire Harness DevOps platform. For experimentation focus, set HARNESS_TOOLSETS=feature-flags so the agent sees just the FME workspaces, environments, flags, and segments. Without scoping the agent gets the full DevOps surface, which crowds tool selection on smaller models.

Official MCPApache 2.0 OSS. Local stdio or HTTP transport.

What you can do via MCP

Example prompts the agent runs.

  • “List all FME workspaces and environments.”

    Returns the FME workspace list with their environments so the agent can route subsequent calls to the right scope.

  • “Create a new feature flag in the production workspace.”

    Creates the flag in the named FME workspace, returns the key, and stages the SDK integration snippet.

  • “Toggle the new-checkout flag in environment staging.”

    Updates flag state in the staging environment via the Harness MCP and confirms the rollout config.

  • “Update targeting rules for the pricing-test flag.”

    Edits the targeting rule list, surfaces the diff, and confirms the change applied to the right environment.

  • “Archive the post-launch flag for the December release.”

    Archives the flag with the right metadata so historical audit trails stay intact and the SDK stops evaluating it.

Split.io (Harness FME) · MCP

Local Harness MCP invocation, scoped to feature flags. The HARNESS_TOOLSETS env var keeps the agent's view focused on FME.

Split.io (Harness FME) MCP
docs
# Harness MCP, scoped to FME toolset
# Run with: HARNESS_API_KEY=pat_abc \
#           HARNESS_TOOLSETS=feature-flags \
#           npx harness-mcp-v2

mcp.harness.create_feature_flag({
  workspace: "production",
  environment: "production",
  name: "new-checkout",
  key: "new_checkout",
  variations: [
    { name: "on", value: true },
    { name: "off", value: false }
  ],
  defaultVariation: "off"
})

# IMPORTANT: omit HARNESS_TOOLSETS to get all 137 resource types
One command sample showing how the agent talks to Split.io (Harness FME). The MCP exposes the platform's primitives as tools; the agent translates the prompt into the right call.

MCP integration

Split.io (Harness FME) MCP server.

Server
npx harness-mcp-v2
Auth
HARNESS_API_KEY (Personal Access Token)
Hosting
Local stdio or HTTP transport
Scoping note
Set HARNESS_TOOLSETS=feature-flags. Without this scope the agent sees 137 resource types across the entire Harness DevOps platform, which crowds tool selection.
  • Apache 2.0 open source
  • Covers FME workspaces, environments, flags, segments, targeting rules
  • Same MCP exposes the rest of Harness when other toolsets are added
  • Pairs with the Harness CLI for build-time and CI flag operations

Harness MCP docs

Visual demonstration

What this looks like in practice.

Split.io (Harness FME)/Feature flag
Release

new_checkout

Single-step checkout flow, FME-managed

Environment

production

Off0% rollout

Targeting rules

  • segment IN ('beta-testers') ENABLE 100%
  • segment IN ('internal-users') ENABLE 100%
  • ELSE off
FME flag managed via the scoped Harness MCP. The agent reads and writes flag state, targeting, and segment membership.

CLI alternative

Harness CLI exists; MCP for AI workflows.

Harness ships a CLI for build-time flag evaluation, CI integration, and bulk operations. For AI-driven feature work the scoped MCP is the right interface; the CLI fills in for non-agent flows.

Pairs with these skills

The PM-experimentation skill suite.

This integration pairs with the forthcoming feature-flagging skill. The skill landing pages and SKILL.md sources land in subsequent dispatches; cross-link hyperlinks are added when the skill pages ship.

Continue reading.