Cursor Token Optimization

In Cursor user plan mode for complex changes. First use PLAN → review the PLAN and make changes if necessary → choose optimal LLM and then hit BUILD, so you don’t spend tokens on unwanted code changes.w

See the Model Guidelines below.

Some models tend to generate excessive documentation, which might not be necessary — address this in your prompt or in rules/commands.

Make sure your Cursor workspace is indexed — you can verify and run indexing manually in Settings → Indexing & Docs.

For documentation, prefer using Context7 MCP: https://upstash.com/blog/new-context7

Sometimes manually changing a couple of lines in a file is more time- and cost-efficient than using the Agent.

Be hyper-specific in your prompts. Specific requests = narrower context = fewer tokens.

Use @mentions to target context.

Close irrelevant open files. Cursor includes open files as automatic context. Close tabs you’re not actively working on.

Use .cursorignore aggressively. Exclude directories that should never be context.

Break large tasks into smaller chunks. Each focused task uses less context and is more token-efficient.

Clear context between unrelated tasks. Start a new chat for unrelated topics instead of continuing in the same thread.

Golden rule: use common sense.

Cursor Token Optimization

# How to optimise tokens

# Resources

Backlinks