IDE context alternative: trace-mcp vs Cursor rules and Copilot instructions

TL;DR. Cursor rules (.cursor/rules with description/globs/alwaysApply, plus team rules and AGENTS.md) and Copilot instructions (the repo-wide copilot-instructions.md file in .github, plus path-specific .instructions.md files under .github/instructions) are the same idea in two IDEs: markdown guidance, versioned with the repo, telling the agent how to work here. trace-mcp is a different layer underneath: a persistent graph of what the code is — symbols, imports, call edges, 88 framework integrations — served over MCP to any agent. Rules steer the session; the graph answers the structural questions inside it.

This is a conceptual comparison — there is no single repository on the other side, so there is no star count and no version. Facts below are from Cursor docs and GitHub docs as read on September 26, 2026.

Head-to-head

Capability trace-mcp IDE rules (Cursor / Copilot)
What it is computed code graph over MCP authored markdown guidance for one agent
Who writes it the indexer (from your code) you (conventions, intent, taste)
Answers who-calls-this ✓ one tool call, stored edges ✗ can instruct the agent to check; holds no edges
Impact analysis ✓ reverse dependency traversal ✗
Framework-aware edges ✓ 88 integrations ✗
Per-file scoping n/a (queries scope themselves) ✓ globs (**/*.py), path-specific instruction files
Team enforcement n/a ✓ team rules, enforced or optional
MCP tools advertised (default) 29 (~11.6K tok); 182 on full n/a — rules are prompt text; MCP servers attach per host
Persists across sessions ✓ SQLite index + code-linked memory ✓ files persist; agent re-reads them per session
Works across agents ✓ any MCP client partial — AGENTS.md travels; .mdc and team rules do not
Refactoring write path ✓ AST rename, move, extract, codemod ✗ (the agent edits; rules only constrain it)
Security scanning ✓ OWASP Top-10 taint, SARIF 2.1.0 ✗
Setup cost index build, then instant queries ~zero — a markdown file

Verified on September 26, 2026 against Cursor rules docs (cursor.com/docs/rules, cursor.com/docs/context/rules) and GitHub Copilot custom-instructions docs plus the VS Code agent-customization page. Cursor rule mechanics (.mdc frontmatter, four rule types, Always/Intelligent/Files/Manual application) and Copilot mechanics (repo-wide file, path-specific .instructions.md, organisation level) are as those pages state; anything beyond them is marked as guidance, not quoted fact.

When to pick IDE rules

Honest version — and most teams should do all of this regardless of the graph:

When to pick trace-mcp

The honest caveat

Rules win on cost and on content no tool can derive: a 200-line markdown file versioned in git beats any index for “how we work here”, and at ~zero tokens of machinery it is the first thing every team should write — before us, not instead of us. The graph is the second layer, not the first.

And our standing ceiling, stated on the comparisons page: line-based CFG, lexical taint with type-aware pruning — not a dataflow engine, and out of scope to become one.

FAQ

What is the difference between IDE rules and a code graph? Authored guidance versus derived structure. Rules tell the agent how to behave here; the graph answers what the code does. One you write, the other the indexer computes.

Do Cursor rules or Copilot instructions answer who-calls-this? No. They can demand the agent check, but they hold no edges — the agent still searches. The graph stores the edges, so the question is one call.

Can trace-mcp replace my Cursor rules? No. Rules carry team judgement the code cannot contain; the graph carries facts the rules cannot contain. Keep the rules; add the graph underneath.

Where does each one live? Rules live with the repo and IDE (.cursor/rules, team dashboard, the Copilot instructions files, AGENTS.md); the graph lives beside the repo as SQLite, served over MCP to any agent.

Next steps

Last updated: September 26, 2026