@splithub/mcp

MCP server for SplitHub. Lets Claude Code, Cursor, Codex or any MCP client run experiments (create, start, pause, finish, read verdicts, ship the winner) and read analytics (traffic, sources, funnels, paths, retention, realtime) through tools, on websites and in apps (React Native, Expo, iOS, Android), with the same config file and credentials as the splithub CLI.

npx splithub skills --mcp      # writes .mcp.json (Claude Code) and .cursor/mcp.json (Cursor)
# or by hand:
claude mcp add splithub -- npx -y @splithub/mcp

{ "mcpServers": { "splithub": { "command": "npx", "args": ["-y", "@splithub/mcp"] } } }

Tools:

| Area | Tools | |---|---| | Experiments | splithub_status, splithub_results, splithub_experiment_events, splithub_plan, splithub_experiment_create, splithub_experiment_status, splithub_refs, splithub_ship, splithub_remove, splithub_review | | Analytics | splithub_analytics, splithub_analytics_connect, splithub_analytics_events, splithub_analytics_funnel, splithub_analytics_funnels (saved funnels), splithub_analytics_paths, splithub_analytics_retention, splithub_analytics_realtime, splithub_segments, splithub_property_create, splithub_property_status | | Around them | splithub_notes, splithub_links, splithub_alerts, splithub_domains | | Project | splithub_init, splithub_push, splithub_pull, splithub_doctor, splithub_login, splithub_whoami, splithub_projects |

Apps:

• splithub_init knows React Native and Expo: it installs @splithub/react-native and wraps the app entry with screen tracking. In a native iOS or Android project it returns the install lines and a quick start.
• splithub_doctor checks the app package and its storage module (or the native SDK in the build files).
• splithub_property_create takes kind: site | app (app by default in an app project).
• Experiment results, breakdowns and every analytics report take platform (ios, android, web…) and app_version as a filter or a by dimension.
• splithub_analytics_retention gives daily or weekly cohorts: who comes back after the first open.

The old names splithub_create_experiment, splithub_set_status, splithub_create_property and splithub_set_property_status still work for one release.

With SplitHub Cloud, results carry the verdict (verdict.message), a friendly line (voice.line) and ready-to-show text (display). Self-hosted projects get raw numbers only (analysis.available: false): the agent reports them and says verdicts come with SplitHub Cloud. splithub_plan asks SplitHub Cloud for the sample size; no account needed.

Env: SPLITHUB_PROJECT_DIR (defaults to the client's cwd), SPLITHUB_API (staging / self-hosted cloud), SPLITHUB_API_KEY (CI), SPLITHUB_CLI (path to a specific CLI build).

The skill .claude/skills/splithub-mcp/SKILL.md (installed by npx splithub skills) tells the agent how and when to use each tool.