## Context This is the announcement and feedback issue for the **Beta Release** of the **Semantic Code Search** feature. This applies to the **[`semantic_code_search` MCP tool](https://docs.gitlab.com/user/gitlab_duo/model_context_protocol/mcp_server/#semantic_code_search)** used in **Agentic Chat**, though you may use other MCP clients (e.g. [Claude Code](https://docs.gitlab.com/user/gitlab_duo/model_context_protocol/mcp_server/#connect-claude-code-to-a-gitlab-mcp-server)) instead of Agentic Chat. With the Beta Release, the `semantic_code_search`: - will be available to **all users that have access to the [Gitlab MCP Server](https://docs.gitlab.com/user/gitlab_duo/model_context_protocol/mcp_server)** - can be invoked for **all projects with [Duo availability turned on](https://docs.gitlab.com/user/gitlab_duo/turn_on_off/#turn-gitlab-duo-on-or-off)** - This requires a project's files to be indexed into Code Embeddings. - If this tool is invoked for projects without Code Embeddings yet, indexing will be triggered ad-hoc and the LLM Agent will be instructed to use a different tool. The semantic search will be available when invoked again after a few minutes. Related issue: https://gitlab.com/gitlab-org/gitlab/-/issues/569746+ ## Setup links - [Agentic Chat](https://docs.gitlab.com/user/gitlab_duo_chat/agentic_chat/) - [Instructions for connecting to any MCP Server from any MCP Client](https://docs.gitlab.com/user/gitlab_duo/model_context_protocol/mcp_clients/) - note: Agentic Chat / Duo Agent Platform is considered an MCP client - [Configuration instructions for connecting to the Gitlab MCP Server](https://docs.gitlab.com/user/gitlab_duo/model_context_protocol/mcp_server/) ## Example Usage Scenarios on Agentic Chat 1. **semantic code search performed on the Workspace's active project** <details> <summary>Expand for screen recording</summary> {width=819 height=600} </details> 2. **semantic code search performed on a project specified in the user question** <details> <summary>Expand for screen recording</summary> {width=819 height=600} </details> 3. **semantic code search performed on a project without Code Embeddings yet** 1. The LLM Agent will try to use other tools to answer the user's question. <details> <summary>Expand for screen recording</summary> {width=819 height=600} </details> 2. Meanwhile, Code Embeddings Indexing will be triggered ad-hoc behind the scenes 3. After 10 - 20 minutes, semantic code search should be available on the project <details> <summary>Click to expand</summary> {width=819 height=600} </details> 4. **semantic code search performed on a project included through `/include repository`** This is not available in the Beta Release. Please see below for further details. ## Excluding `/include repository` for the initial Beta Release The `/include repository` command is used by both the **Agentic Chat** and **Classic Chat**. Given that **[Classic Chat is already Generally Available](https://docs.gitlab.com/user/gitlab_duo_chat/)**, we cannot enable this command without making it generally available to all users using Classic Chat. Additionally, this command has a pending improvement issue required for it to list projects that have _Duo availability turned on_ (see https://gitlab.com/gitlab-org/gitlab/-/issues/576635 & https://gitlab.com/gitlab-org/editor-extensions/gitlab-lsp/-/issues/1629) For these reasons, we will skip the `/include repository` command in the Beta Release. Instead, it will be rolled out as **General Availability release for Classic Chat** and a **follow-up to the Beta Release for Agentic Chat**. ### Example Usage Scenarios of the `/include repository` command 1. **AGENTIC CHAT:** `semantic_code_search` performed on a project specified through `/include repository` <details> <summary>Expand for screen recording</summary> {width=819 height=600} </details> 1. **CLASSIC CHAT:** semantic search performed on a project specified through `/include repository` Selecting a project through `/include repository` is the only way semantic search can be performed on Classic Chat. This is different from Agentic Chat, where the LLM can perform semantic search depending on the user prompt. <details> <summary>Expand for screen recording</summary> {width=819 height=600} </details>