Manage Model Context Protocol (MCP) servers across all supported agent runtimes

[dnviti]

Problem / Motivation

Users running coding agents often rely on Model Context Protocol (MCP) servers to provide external tools, documentation, database connections, and custom API capabilities. Currently, configuring MCP servers across different supported agents requires manually editing agent-specific configuration files or passing environment variables. Because each supported agent runtime handles MCP configurations differently (or lacks native file configuration entirely), users face inconsistent setup workflows, broken tool integrations, and lack a single dashboard to view, test, and toggle active MCP tools across their active coding sessions.

Goal

Provide a unified, cross-runtime MCP server management interface and sync system that allows users to define global and project-scoped MCP servers once and seamlessly toggle, test, and apply them across all supported coding agent runtimes.

Scope

• Centralized MCP Server Catalog: Support creating, viewing, updating, and removing MCP server definitions at both the global (user-scoped) and project (workspace-scoped) levels.
• Supported Transports: Support standard MCP command-line (stdio with command, arguments, and environment variables) and remote network (SSE/HTTP endpoints) server definitions.
• Per-Runtime Enable/Disable Toggles: Allow users to enable or disable individual MCP servers globally or selectively for specific agent runtimes.
• Connection Testing & Status Monitoring: Include a diagnostic action to test connectivity to an MCP server and report clear status indicators (e.g. Connected, Degraded, Disconnected, or Untested).
• Cross-Runtime Provisioning: Automatically propagate enabled MCP server configurations to agents with native configuration mechanisms, and provide runtime bridge integrations for agents without native MCP file configuration.
• Access Control & Permissions: Restrict global/host-wide MCP server modifications to the installer account while allowing any signed-in user to manage private project and session MCP servers.

Non-goals

• Implementing custom internal MCP tools or authoring new MCP servers within the application.
• Managing third-party API keys or secret stores outside standard MCP environment variable inputs.
• Modifying underlying system binaries or OS-level package managers to install MCP dependencies automatically.
• Automatic auto-discovery or scanning of untrusted local ports for unlisted MCP servers.

Security / Safety Considerations

• Executing stdio MCP server commands runs local processes with the user's host/container execution permissions; server definitions must require explicit user creation or approval.
• Environment variables containing sensitive credentials (such as API keys) associated with MCP servers must remain encrypted at rest and restricted to authorized user sessions.
• Modifying global host-wide MCP configurations remains restricted to the installer account to prevent unauthorized privilege escalation across shared server sessions.

Acceptance Criteria

• A dedicated MCP server management dialog is accessible from global application settings, runtime launcher profiles, and project settings.
• Users can add, edit, duplicate, and delete stdio and SSE/HTTP MCP server definitions.
• Users can toggle an MCP server on or off overall, or select specific target agent runtimes where the server should be active.
• Each MCP server entry includes an interactive "Test Connection" action that displays a human-readable connectivity result without interrupting active agent sessions.
• Starting a new session with any supported agent runtime automatically provisions enabled MCP servers for that runtime environment.
• Updating an MCP server definition updates connected active sessions or prompts the user with clear restart guidance when an agent runtime requires a session restart to pick up configuration changes.
• Project-scoped MCP server configurations persist with project settings and override or complement global MCP server definitions cleanly.
• Non-installer accounts can manage project and session-scoped MCP servers but are prevented from modifying global host-shared MCP configurations with an explanation.
• A comprehensive product demonstration verifies adding a stdio MCP server, testing connection status, enabling it across multiple runtimes, and launching an agent session with verified tool access.
• User documentation reflects MCP server management workflows, transport requirements, and per-runtime compatibility details.

Suggested Phasing

1. Phase 1: Core MCP catalog storage, authorization model, and centralized management UI with stdio and SSE/HTTP server definition support.
2. Phase 2: Interactive connection testing, per-runtime toggle matrix, and native configuration synchronization.
3. Phase 3: Extension bridge wrappers for non-native runtimes, project-scoped overrides, in-session restart notices, and user documentation.