Compatibility

Works with every major
AI coding environment.

One command. Strata auto-detects your installed IDEs, registers the MCP server in each one, and writes the right instructions file for each format. No model to configure, no key to add.

8
supported IDEs
6
LLM providers
1
command to install
0
manual config required

Auto-registers in all of them.

Running strata install detects which IDEs are present and writes the correct config format for each — no manual editing needed.

Claude Code

Anthropic's official CLI. Detects ~/.claude directory.

~/.claude/settings.json
key: mcpServers
instructions → CLAUDE.md
Cursor

The AI-native editor. Detects ~/.cursor directory.

~/.cursor/mcp.json
key: mcpServers
instructions → .cursorrules
Windsurf

Codeium's agentic IDE. Detects ~/.codeium/windsurf.

~/.codeium/windsurf/mcp_config.json
key: mcpServers
instructions → .windsurfrules
VS Code

With GitHub Copilot. Detects the VS Code user config directory.

Code/User/mcp.json
key: servers + type: stdio
instructions → .github/copilot-instructions.md
Zed

The fast GPU-accelerated editor with native MCP. Detects ~/.config/zed.

~/.config/zed/settings.json
key: context_servers
instructions → MCP only
Continue.dev

Open-source AI coding assistant for VS Code and JetBrains. Detects ~/.continue.

~/.continue/config.json
key: mcpServers (array)
instructions → MCP only
Cline

Autonomous coding agent for VS Code. Detects its VS Code extension storage.

globalStorage/saoudrizwan.../cline_mcp_settings.json
instructions → MCP only
Roo Code

Cline fork with expanded model support. Detects its VS Code extension storage.

globalStorage/rooveterinaryinc.../cline_mcp_settings.json
instructions → MCP only
How auto-detection works. Strata checks for each IDE's config directory at install time. Any IDE whose directory exists gets registered. Pass --claude, --cursor, --zed, --cline, etc. to target specific IDEs, or --all to force all of them.

There is no provider to bring.

Selection, composition and delivery all run in Node, before your agent sees a byte. The only model involved is the one you are already talking to.

Zero LLM calls

Nothing is generated on the fly inside Strata, so nothing can be hallucinated. The modules are written and tested ahead of time; the pipeline only selects and composes them.

Zero keys, zero cost

No account, no API key. Strata sends the task it was given — never your source, your schema or your files.

Your agent names the parts

Your model names the parts of the job — it has read the whole task, so it knows what the work is made of. Strata scores those names against the library.

Why deterministic? Because the expensive part of an agent session is not writing a feature — it is proving the feature works. A pipeline that adds its own LLM call adds its own latency, its own cost, and its own chance to be wrong. Strata spends none of those, and hands your agent a verifier instead.
One command

Everything configured in seconds.

Install Strata globally. It registers itself in every IDE it finds, writes the right instructions file for each, and prints exactly what it did.

# Install globally
npm install -g strata-lib

# Auto-detects Claude Code, Cursor, Windsurf, VS Code, Zed, Continue, Cline, Roo
✓ Registered Claude Code (~/.claude/settings.json)
✓ Registered Cursor (~/.cursor/mcp.json)
✓ Created CLAUDE.md
✓ Created .cursorrules

# No provider to set, no key to add — it just runs.

# Target specific IDEs manually
strata install --zed --cline --roo
Install