# Welcome to the MCP servers in AI Catalog feedback issue > [!NOTE] Feature maturity status > Experimental The purpose of this feedback issue is to collect your experiences with external MCP support in the AI Catalog. Broad feedback ranging from registering and managing servers (self-managed customers only), to connecting them to agents and using them in practice. Our goal is to understand what's working, what's broken, and what's missing so we can continue to mature this feature toward beta and GA. <details> <summary>What is MCP server support in the AI Catalog?</summary> The AI Catalog now supports connecting Duo Agent Platform's custom agents to external third-party services via Model Context Protocol (MCP). This means your agents can reach beyond GitLab's built-in tools to interact with services like Jira, Datadog, Snyk, and more — directly from the agent configuration UI. This is the **GitLab agent → external service** direction of MCP: your agents are the clients, and the MCP servers you configure are the data sources and action providers. For customers on: - **GitLab.com:** A curated set of MCP servers has been seeded into the global catalog for you. See the roadmap of available servers we will be adding [here](#591969). - **Self-managed:** Admins can register MCP servers directly in their instance's AI Catalog under **AI Catalog > MCP**. ([docs](https://docs.gitlab.com/user/gitlab_duo/model_context_protocol/ai_catalog_mcp_servers/#add-an-mcp-server-to-the-ai-catalog)) </details> <details> <summary>What's in scope for this experimental release?</summary> - Registering and managing MCP servers in the AI Catalog _(self-managed customers only)_ - Curated server list seeded and managed by GitLab _(GitLab.com customers only) - Associating MCP servers with custom agents via the agent edit form - OAuth authentication with supported MCP servers - HTTP transport only What's **not** yet supported: tool-level restrictions within a server, local MCP servers, and governance/observability controls. These are on our radar for [future iterations](https://gitlab.com/gitlab-org/gitlab/-/work_items/571237). </details> ## :dart: Feedback we're especially interested in This is an early experimental release and your input will directly shape what we prioritize next. #### For GitLab.com customers — MCP server coverage We've seeded the catalog with an initial set of servers across issue tracking, cloud infrastructure, security, observability, and data (see full list [here](#591969)). We want to know: 1. **What's missing?** Which MCP servers are critical to your workflow that aren't in the catalog yet? 2. **What's not working?** Are any of the seeded servers behaving unexpectedly or failing to connect? #### For self-managed admins — the setup and management experience 1. **How was the registration process?** Was adding MCP servers to your instance catalog clear and manageable, or did you hit friction? 2. **What do you need to manage servers at scale?** Think bulk registration, environment-specific configs, or syncing across instances. 3. **What's missing from the admin experience** that would give you confidence to roll this out broadly? #### For everyone — governance and observability We know enterprise customers need visibility and control over how agents interact with external systems. Help us understand your requirements: 1. **Governance:** What controls do you need over which MCP servers agents can access? Do you need to restrict access by user, group, or project? 2. **Observability:** What visibility do you need into agent ↔ MCP server interactions? Think audit logs, call history, error rates, or cost. 3. **Trust and vetting:** What does "trusted MCP server" mean to you? What signals or certifications would give you confidence to enable a server for your org? ## :pencil: How to give feedback 1. **Check existing threads:** Review comments below to see if your issue is already reported. Add a :thumbsup: or comment to show support. 2. **Start a new thread:** Use a descriptive title so others can quickly understand the focus of your thread. 3. **Include context:** Tell us whether you're on GitLab.com or self-managed, which MCP server you were working with, what you were trying to do, and what happened. Screenshots or error messages are especially helpful. ## :handshake: What you can expect from us 1. We **will read** all feedback while this issue remains open 2. We **will prioritize** fixes and iterations based on feedback patterns 3. We **will create issues** for reproducible bugs 4. We **may reach out** to invite you into design partnership conversations as we plan the next iteration --- :clap: Thank you for helping us build a better agent platform. Your feedback directly shapes what we build next.