Skip to content

Use Canopy Alongside GitHub Copilot

GitHub Copilot and Canopy solve different problems. Copilot generates code completions as you type. Canopy provides structural intelligence: dependency graphs, health checks, symbol search, and pre-modification impact analysis. They don’t conflict — they complement each other.

What each tool does

Section titled “What each tool does”

GitHub Copilot:

  • Autocompletes the line you’re typing
  • Suggests function bodies based on surrounding context
  • Generates boilerplate from a comment or type signature
  • Works in the editor inline, without switching context

Canopy (via MCP):

  • Answers “what depends on this file?” before you rename a function
  • Finds all callers of an interface before you change its signature
  • Surfaces broken imports and dead exports before they reach CI
  • Searches for all usages of a pattern across the entire codebase
  • Provides a GO/CAUTION/STOP assessment before a refactor begins

They operate at different levels: Copilot is line-level, Canopy is codebase-level.

Section titled “Recommended setup”

Use Copilot for inline completions in VS Code or any editor with the Copilot extension. Use Canopy via an MCP-connected AI assistant — Claude Code, Cursor, or Continue — for the planning and analysis phase of work.

A typical workflow:

  1. Before a refactor, ask your AI assistant (Claude Code, Cursor, Continue) to call canopy_prepare on the file. Review the dependency count and health findings.
  2. Plan the refactor with the AI assistant, informed by Canopy’s structural data.
  3. Implement using Copilot’s inline completions for the actual code edits.
  4. After the refactor, ask the AI assistant to call canopy_validate on changed files.

This is not a rigid ceremony — skip Canopy for trivial edits (typo fixes, documentation changes). Use it for anything that touches an exported interface, moves a file, or restructures shared code.

No conflict with Copilot

Section titled “No conflict with Copilot”

Canopy runs as an MCP server — it has no VS Code extension, no autocomplete integration, and no inline suggestions. It doesn’t compete with Copilot for the completion popup or interfere with Copilot’s inference. You can have both active simultaneously without any interaction between them.

If you’re using Copilot in VS Code, you can also use Canopy via:

  • VS Code + Continue — Continue is an AI chat panel that supports MCP
  • Claude Code (separate terminal) — run alongside VS Code
  • Cursor (separate window) — Cursor supports both Copilot and MCP simultaneously
Section titled “Copilot Chat vs. Canopy search”

Copilot Chat (the / command panel in VS Code) can answer questions about your codebase using its own index. For questions like “what files import this component?” Copilot Chat may answer — but it works from a probabilistic model of the code, not a precise dependency graph.

Canopy answers the same question with a deterministic graph traversal. If you need precision (“give me every file that will break if I delete this export”), use Canopy. If you want a rough sketch or a natural language explanation, Copilot Chat is faster.

The two are complementary here too. Ask Canopy for the precise list, then paste it into Copilot Chat and ask for an explanation or a refactoring plan.

Workflow example

Section titled “Workflow example”

Scenario: Rename createPayment to initiatePayment in src/lib/payments.ts.

Step 1 — Canopy analysis (via AI assistant):

Call canopy_prepare on src/lib/payments.ts and tell me how many files import it
and whether there are any P0 health issues.

Output: 23 dependents, no P0 findings, assessment is CAUTION.

Step 2 — Plan with AI assistant:

I want to rename the export 'createPayment' to 'initiatePayment'. Based on the
23 dependents you found, what's the safest approach?

The AI assistant — informed by Canopy’s dependency data — recommends updating all call sites atomically.

Step 3 — Implement with Copilot: Open each dependent file. Use Copilot’s completions to make the rename changes quickly. Copilot learns from the pattern you’ve established in earlier files.

Step 4 — Validate with Canopy:

Call canopy_validate on src/lib/payments.ts and the 5 files we just edited.

Canopy confirms no new broken imports were introduced.

Common pitfalls

Section titled “Common pitfalls”

Asking Copilot Chat for dependency information Copilot Chat’s answers about “what imports this file” are approximate — it works from its training data and a snapshot of your open files, not a real-time graph. For accurate dependency counts before a breaking change, use canopy_trace_dependents.

Running canopy serve in VS Code terminal while Copilot is active No conflict. canopy serve is a background process — it doesn’t interact with VS Code’s extension layer. You can leave it running in the integrated terminal without any effect on Copilot.

Section titled “Related guides”