MCP Overview

The Hyphen MCP server connects AI assistants and code agents to your organization. Claude, Cursor, and VS Code can manage feature flags, deployments, ENV metadata, Agent tasks, and audit events. Each request uses your Hyphen App permissions.

MCP (Model Context Protocol) is the open standard AI clients use to call external tools. Hyphen hosts an MCP server, so there is nothing to install or run yourself:

The server is authenticated by default: every request needs a Hyphen credential, and your organization is resolved automatically from it. If you belong to multiple organizations, pin one with the /<organization id>/ form of the URL.

Table of Contents

• What You Can Do
• Access and Safety
• Health Endpoints
• Next Steps

What You Can Do

The server provides tools across the Hyphen platform. A connected agent can use these tools through conversation:

• Ship code — "Deploy this repo to a preview environment" walks through project and app setup, Docker packaging, environment variables, the deployment run, and its logs. See Deploy.
• Manage feature flags — create and update toggles and segments, then verify with a live edge evaluation that returns the same answer a real user would receive.
• Clean up stale flags — telemetry-backed tools find flags that no longer vary, and a guided workflow removes them from your codebase before retiring them in Hyphen.
• Investigate incidents — search the organization audit log and correlate it with deployment runs, logs, and metrics to answer "what changed right before things broke?"
• Operate Agent tasks — schedule and inspect Hyphen Agent automations, review their runs, and answer their questions when a run is waiting on input.
• Answer platform questions — search these docs live, so answers come from current documentation instead of stale training data.

See the tool reference for documented tools. See prompts and resources for workflows.

Access and Safety

• Your credential, your permissions. Every tool call runs as you. What a connected agent may do is decided by the Hyphen API per request from your organization membership and roles — the same rules as the Hyphen App. Sign in with OAuth or an API key; see Getting Connected.
• Permission prompts on writes. Tools that modify data are annotated as such, and destructive operations (deletes, deployment run triggers, agent chat) carry an additional destructive hint, so well-behaved MCP clients ask for your confirmation before running them.
• A production gate on deployments. A deployment run that targets a production-type environment is refused unless the call explicitly acknowledges production — and agents are instructed to do that only after you confirm.
• ENV secret values never pass through the server. Hyphen ENV values are end-to-end encrypted with a key only you hold, so the server can only ever read secrets metadata (names, counts, versions). Use the Hyphen CLI (hx env pull) locally when you need values.

Health Endpoints

Two public, unauthenticated endpoints let uptime monitors probe the server: GET /pingz for liveness and GET /healthz for a health report with per-dependency detail (200 when healthy, 503 when any dependency is down).

Next Steps

• Getting Connected — authenticate and set up your MCP client
• Tool Reference — documented tools by product area
• Prompts and Resources — packaged workflows like deploy_app and cleanup_stale_flags

The server is open source at github.com/Hyphen/mcp and published to the official MCP registry as ai.hyphen/mcp.