# Expose LSP Find All References to the Agent

**URL:** <https://forum.cursor.com/t/expose-lsp-find-all-references-to-the-agent/170868>\
**Category:** Feature Requests\
**Tags:** language-servers, context\
**Created:** [September 7, 2026, 9:09am UTC](https://forum.cursor.com/t/expose-lsp-find-all-references-to-the-agent/170868 "2026-09-07T09:09:16Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![LarrySmithIDIC](https://sea3.discourse-cdn.com/cursor1/user_avatar/forum.cursor.com/larrysmithidic/32/120114_2.png) [@LarrySmithIDIC](https://forum.cursor.com/u/LarrySmithIDIC)\
**Post date:** [September 7, 2026, 9:09am UTC](https://forum.cursor.com/t/expose-lsp-find-all-references-to-the-agent/170868/1 "2026-09-07T09:09:16Z")

</div>

### Feature request for product/service

Cursor IDE

### Describe the request

The Agent already consumes LSP diagnostics (the same store as the Problems panel). It does not get other language-server results that the IDE already computes.

In the C# extension, CodeLens shows “1 reference” on a method. Clicking it opens Peek References (LSP textDocument/references) with the call sites plus the definition. That is the same graph a human uses to judge blast radius and which tests to run.

The Agent cannot query that. It approximates with text search (grep on the identifier). That misses interface/base-class call sites, overloads, and generated code, and it over-matches comments and unrelated types. For a private method like ParseExtensions that is cheap; for Parse / Read / a public interface member it is slow and wrong.

Ask: give the Agent a first-party tool that calls the workspace language server, starting with:

1. Find all references — input: file path + line/column (or fully qualified name); output: the same location list Peek References shows
2. Later: go to definition / go to implementation
3. Later: rename symbol (workspace edit), instead of rewriting files by string replace

Diagnostics already flow to the Agent. References are the next high-value slice of the same protocol. This would speed test selection, refactors, and “what calls this?” questions without standing up a second compiler workspace.

Related existing request: [Support of LSP (Language Server Protocol)](https://forum.cursor.com/t/support-of-lsp-language-server-protocol/145628)

### Operating System (if it applies)

Windows 10/11

---

<div class="post-metadata">

**Author:** ![Colin](https://sea3.discourse-cdn.com/cursor1/user_avatar/forum.cursor.com/colin/32/90452_2.png) [@Colin](https://forum.cursor.com/u/Colin)\
**Post date:** [September 7, 2026, 11:41am UTC](https://forum.cursor.com/t/expose-lsp-find-all-references-to-the-agent/170868/5 "2026-09-07T11:41:19Z")

</div>

Hi @LarrySmithIDIC, thanks for the suggestion!

This isn’t on our short-term roadmap, but in the meantime you may want to explore tools that integrate with Cursor, such as the [Serena MCP server](https://github.com/oraios/serena).