AgentsMedium impactFor DevGitHub MCP Servers · May 23, 2026
Agent profile manager for Claude Code and Codex. Per-directory profiles select which skills, MCP servers, and Claude Code plugins load - automatically, before launch.
recodeee/cue
Recodeee/cue is a TypeScript agent profile manager that automatically selects skills, MCP servers, and Claude Code plugins per directory before agent launch.
Signal strength3.8/5·1 stars
Recodeee/cue is a TypeScript agent profile manager that automatically selects skills, MCP servers, and Claude Code plugins per directory before agent launch.
TL;DR
Recodeee/cue is a TypeScript agent profile manager that automatically selects skills, MCP servers, and Claude Code plugins per directory before agent launch.
What happened
A new open-source tool called cue was released to manage agent profiles for Claude Code and Codex, enabling automatic configuration of agent capabilities and plugins based on directory context.
Why it matters
It streamlines development workflows by dynamically customizing AI agent environments, improving efficiency and enabling more modular agent setups.
Generating deep dive...
AI-powered analysis takes a few seconds
The bigger picture
The introduction of cue illustrates a maturing AI developer ecosystem that demands modularity and automation akin to traditional software engineering practices. As AI agents become embedded across diverse workflows, the ability to contextually toggle capabilities per project mirrors the evolution of environment managers for programming languages or container orchestrators for applications. It signifies a trend toward treating AI agent configurations as first-class artifacts within developer environments, which enables scalable and maintainable AI workflows. Moreover, cue's approach of integrating with multiple agent platforms like Claude Code and Codex suggests pressure toward platform-agnostic tooling. This reflects an industry grappling with fragmentation and seeking unified ways to manage AI components dynamically. Overall, cue is a small but telling signal that successful AI adoption hinges not only on model capabilities but also on seamless developer ergonomics that reduce setup friction.
Technical deep dive
Cue is architected as a TypeScript agent profile manager that intercepts the agent startup process to inject configuration context based on the active directory. It operates by parsing structured profile files-likely JSON or YAML-that specify which set of skills, MCP servers, and Claude Code plugins to include. This avoids the common pattern of environment-wide static configurations, enabling granular control at the directory level. Under the hood, cue must interact with agent launching scripts or CLI wrappers to override default plugin registries before the runtime initializes. This implies integration points both at the filesystem event level and within the agent framework’s plugin loader. The design choice to support both Claude Code and Codex broadens applicability but creates challenges around normalization of plugin interfaces and version compatibility. Strategically, investing in a profile-based declarative system opens pathways for future enhancements such as conditional loading rules, dependency resolution between skills, and user-level overrides. From an engineering perspective, ensuring low overhead and robustness during the agent start phase is critical to maintain developer productivity.
Real-world applications
1
A developer working on multiple AI-assisted code bases can use cue to automatically load language-specific analysis skills and relevant plugins when switching between project directories.
2
An AI research team running experiments with different Claude Code plugin sets per project can enforce consistent and reproducible agent environments without manual adjustment.
3
A consultancy managing client projects can organize unique agent profiles to include client-specific MCP servers and capabilities, preventing cross-contamination of sensitive configurations.
4
DevOps engineers can integrate cue within CI pipelines to ensure agent skill sets align exactly with production environment contexts, reducing deployment surprises.
What to do now
Review existing multi-agent or multi-project workflows to identify areas where manual skill and plugin management introduces friction or errors.
Experiment with cue by creating per-directory profiles for a subset of projects and validate that agents load the correct skills and plugins automatically.
Build internal documentation and best practices around profile configuration, emphasizing version control and environment consistency.
Engage with the cue open-source community to contribute feedback, report edge cases, and explore integrations with other AI agent platforms.