Configure your ecosystem
← Set up your repository · Back to the guide · Next: Tips & troubleshooting →
Rosetta tells the agent how to work. CLIs and MCP servers give it the eyes and hands to do that work — reading tickets, driving a browser, fetching current library docs. What you need depends on which scenarios you run.
Prefer CLIs
Prefer a CLI over the equivalent MCP where one exists — CLIs cost no context and are always available. Keep at most three MCPs enabled at a time.
gh— pull requests, issues, releases, CI checksacli— Jira and Confluence from the terminalrg— ripgrep; most current models prefer it over grep
Recommended MCP servers
Bold entries are strongly recommended. The rest depend on your project.
- Context7 — up-to-date library documentation
- Playwright MCP — interact with web pages through structured accessibility snapshots
- Chrome DevTools — full browser control with console, network tab, snapshots
- Graphify — MIT-licensed knowledge graph of your project. Third-party tool with access to your IP; review with your manager
- Jira & Confluence MCP — tickets, comments, and documentation
- GitNexus — indexes a codebase into a knowledge graph. Third-party tool with access to your IP. Free for non-commercial or personal use, paid for commercial use — review the licence and your company policy with your manager
- Figma MCP — so the agent can see designs directly
- Repomix MCP — packages a library into reference material the agent can search
- DeepWiki — up-to-date documentation
- Fetch — retrieve and process content from APIs and web pages
- Database MCPs — read schema, read data
Use either Playwright or Chrome DevTools, not both.
What each scenario needs
| Scenario | Needs |
|---|---|
| Generate test cases | Issue tracker (Jira); test management system to export to |
| Automate API tests | A Swagger/OpenAPI spec or a backend source path |
| Automate UI tests | Playwright or Chrome DevTools to drive the browser |
| Research a question | Context7 and DeepWiki for grounded external references |
| Onboard a library | Repomix (MCP or CLI) |
| Analyze a codebase | Optional: Graphify or GitNexus for a code graph |
Everything else runs on the plugin alone.
Reference code the agent can’t otherwise see
Clone read-only code into refsrc/ as its own subfolder — backend code when this is a frontend repo, corporate or private libraries, or a public framework that had a breaking change in the last year.
Add these exceptions to your root .gitignore:
agents/TEMP/
refsrc/
!refsrc/INDEX.md
Then describe each folder in refsrc/INDEX.md, one header per entry:
## "refsrc/fastmcp-3.3.1" - main framework for MCP handling
## "refsrc/private-ui-lib" - must use corporate styles for TailwindCSS
Reusable patterns
List the patterns the agent should reuse so generated code stays consistent — components, state management, databases, API protocols, messaging, controllers, CRUD verticals.
Project-specific rules
Add your own rules alongside Rosetta — you never edit Rosetta’s files.
| IDE / Agent | Core rules file | Additional rules |
|---|---|---|
| Claude Code | CLAUDE.md |
.claude/rules/*.md |
| Cursor | .cursor/rules/agents.mdc |
.cursor/rules/*.mdc |
| GitHub Copilot | .github/copilot-instructions.md |
|
| Windsurf | .windsurf/rules/*.md |
all .md files auto-load |
| JetBrains (Junie + AI Assistant) | .aiassistant/rules/agents.md |
.junie/guidelines.md |
| Antigravity | .agents/rules/agents.md |
.agents/rules/*.md |
| OpenCode | AGENTS.md |
.opencode/agent/*.md |
Next step
The full workspace checklist, including modernization setup, is in CONFIGURATION. Otherwise pick a task from Scenarios at a glance, or read Tips & troubleshooting →.