diff --git a/roadmap.md b/roadmap.md
new file mode 100644
index 00000000..207a8ac9
--- /dev/null
+++ b/roadmap.md
@@ -0,0 +1,51 @@
+# MCP Inspector Roadmap
+
+This preliminary draft is to collect categories of improvements and discuss them with the maintainers group.
+
+Next steps after deciding on what is highest priority would be to set up a simple Project in the repo and reference the exact issues there.
+
+(Note that these aren't in any priority order yet, just collecting everything in one place.)
+
+## CLI Mode
+
+- Configuration parity with UI mode
+- Feature parity with UI mode
+- Would separating the CLI out into a dedicated repo help with maintainability?
+
+## UX Improvements (UI Mode)
+
+- Not all configs are saved consistently
+- Configs file support is inconsistent between modes
+- Ability to edit/add arbitrary headers
+- Editing long sets of args (or other long values) in the sidebar text boxes is annoying
+- Revisit UI design
+- Server selection and management UI - Surface config file in a chooser to allow closing one server and starting another from the UI
+
+## OAuth Debugging and Compatibility Checking
+
+- Outh testing and debugging hiccups: see issues labeled `auth` for examples
+- Client auth compatibility checker (See: https://github.com/modelcontextprotocol/inspector/pull/694/files)
+
+## Automation/General Process Improvements
+
+- Automatic dependency updates with Dependabot (SDK, etc.)
+- Automation with Claude Code Github Action:
+  - Initial pass at PR review
+  - Initial pass at issue triage/labeling
+  - Opening PRs for simple bugs
+  - Needs references to MCP spec/docs and example tools for testing. Examples that could be incorporated into a 'Maintainer Toolkit MCP Server' or something similar:
+    - https://github.com/olaservo/mcp-advisor
+    - https://github.com/olaservo/mcp-misc/tree/main/example-tools
+- Revisit how we want to label issues and PRs
+- Should we be actively taking thumbs up/voting into account for feature requests?
+- Playwright tests vs ad-hoc testing - what makes sense to expand here as part of CI vs doing testing on-demand?
+  - And what can Claude be instructed to do with Playwright, etc?
+
+## LLM Integration
+
+We previously decided not to include this, but people are finding it useful in alternative testing apps. Our previous concerns were around maintainability and whether this really belongs in Inspector.
+
+Examples:
+
+- Chat functionality to test MCP features
+- Real sampling with an LLM