| name | mcp-use |
| description | Proactively discover and use MCP servers and MCP tools before native CLI tools. Use when searching, reading, editing, refactoring, planning, or researching so Copilot gets higher-signal results with less context waste. |
| allowed-tools | Read, Write, Edit, Glob, Grep |
MCP Use
Use MCP servers and MCP tools first. Prefer MCP-native search, analysis, and editing over shell-builtins or generic CLI commands whenever an MCP tool can do the job more safely or precisely.
When to Use
- Starting any non-trivial task
- Discovering available MCP servers and tools
- Searching code, symbols, files, or external docs
- Reading or editing files with precision
- Structural refactors and code rewrites
- Multi-step planning, debugging, or investigation
- Research that benefits from GitHub or web context
MCP-First Rule
- Check which MCP servers and tools are available.
- Pick the most specific MCP tool for the task.
- Fall back to native tools only when no MCP tool fits or when the MCP tool cannot complete the task.
- Prefer structured MCP output over raw shell output to reduce noise and mistakes.
Tool Selection
- Structural code search: prefer
ast-grep/find_code over plain text grep/rg-only searches.
- Structural refactor: prefer
ast-grep/rewrite_code over manual sed/search-replace.
- Static issue scan: prefer
ast-grep/scan-code over manual spot checks.
- GitHub code research: prefer
octocode/githubSearchCode over broad web search or raw gh search code.
- Read repo files: prefer
fast-filesystem/fast_read_file over cat, sed, or large raw reads.
- Write new files: prefer
fast-filesystem/fast_write_file over shell heredocs for tracked files.
- Precise edits: prefer
fast-filesystem/fast_edit_block over fragile manual replace flows.
- Batch file ops: prefer
fast-filesystem/fast_batch_file_operations over repeated cp/mv/rm commands.
- Complex reasoning: prefer
sequential-thinking/sequentialthinking over jumping straight into edits.
- Web research: prefer
exa/web_search_advanced_exa over generic search first.
Workflows
1) Start by checking MCP options
- Before using Bash or native editor flows, look for a matching MCP tool.
- If a task combines search + edit + reasoning, keep each step MCP-first.
- Prefer the highest-signal tool, not the most familiar one.
2) Search semantically or structurally
- Use
ast-grep/find_code for AST-aware code matches.
- Use
octocode/githubSearchCode for cross-repo or GitHub-hosted code examples.
- Use
exa/web_search_advanced_exa for current docs, release notes, and external references.
3) Read and edit with filesystem MCP tools
- Use
fast-filesystem/fast_read_file to inspect exact file ranges.
- Use
fast-filesystem/fast_edit_block for surgical replacements in existing files.
- Use
fast-filesystem/fast_write_file for new files or full rewrites.
- Use
fast-filesystem/fast_batch_file_operations for coordinated copy/move/delete work.
4) Refactor with AST-aware tooling
- Run
ast-grep/scan-code before or after edits to catch common issues.
- Use
ast-grep/rewrite_code for repetitive transformations across files.
- Prefer AST rewrites over regex replacements when syntax matters.
5) Think before acting on multi-step tasks
- Use
sequential-thinking/sequentialthinking for planning, tradeoffs, or debugging with uncertainty.
- Use it before large edits, workflow redesigns, or cross-file migrations.
Examples
Example: inspect code patterns
Instead of:
rg "TODO|FIXME" .
Prefer:
ast-grep/find_code -> search for syntax-aware patterns in the target language
fast-filesystem/fast_read_file -> inspect only the matching file ranges
Example: perform a safe refactor
Instead of:
rg "oldFunction" . && sed -i ...
Prefer:
ast-grep/find_code -> locate exact call sites
ast-grep/rewrite_code -> apply the structural replacement
ast-grep/scan-code -> check for follow-up issues
Example: edit a tracked file
Instead of:
cat file && python script.py && mv tmp file
Prefer:
fast-filesystem/fast_read_file -> read the current content
fast-filesystem/fast_edit_block -> patch a precise block
fast-filesystem/fast_write_file -> write a new file when needed
Example: research a library or pattern
Instead of:
generic web search + manual browsing
Prefer:
exa/web_search_advanced_exa -> find current docs/articles
octocode/githubSearchCode -> inspect real GitHub usage patterns
sequential-thinking/sequentialthinking -> compare options before implementation
Example: coordinate multiple file operations
Instead of:
cp file1 file2 && mv file3 dir/ && rm file4
Prefer:
fast-filesystem/fast_batch_file_operations -> execute the full batch safely
Decision Rule
If an MCP tool can search, read, reason, rewrite, or research more precisely than a native tool, use the MCP tool first.