IDE Integration
Super-Brain plugs into 14 AI coding tools. Each has a one-command installer that drops the right files in the right places — no manual JSON editing required.
What integration looks like
Super-Brain uses one of three integration styles depending on what the target tool supports:
| Style | Example | How it works |
|---|---|---|
| MCP server | Aider, Claude Code, Gemini | Tool calls Super-Brain over stdio JSON-RPC at query time |
| Hook | Claude Code, Gemini CLI | Pre-tool hook injects Super-Brain context before the assistant runs file searches |
| Rules / skills | Cursor, Copilot, Kiro, Hermes | A markdown rule file the assistant reads on every request |
Most installers use a combination — Claude Code gets a hook and a rule file; Aider gets MCP and an AGENTS.md; Cursor gets a rules file only.
One-command install
All at once
bash
agsuperbrain install --platform all
Per platform
bash
agsuperbrain claude-install # Claude Code
agsuperbrain cursor-install # Cursor
agsuperbrain aider-install # Aider
agsuperbrain codex-install # Codex
agsuperbrain opencode-install # OpenCode
agsuperbrain copilot-install # GitHub Copilot CLI
agsuperbrain vscode-install # VS Code Copilot Chat
agsuperbrain gemini-install # Gemini CLI
agsuperbrain hermes-install # Hermes
agsuperbrain kiro-install # Kiro
agsuperbrain antigravity-install # Google Antigravity
agsuperbrain openclaw-install # OpenClaw
agsuperbrain droid-install # Factory Droid
agsuperbrain trae-install # Trae / Trae CN
Before running any of these, finish the one-time setup:
bash
agsuperbrain init
agsuperbrain ingest ./src
agsuperbrain index-vectors
The graph needs to exist before your assistant can query it.
Platform matrix
| Platform | Install command | Files created | Integration style |
|---|---|---|---|
| Claude Code | claude-install |
.claude/settings.json, CLAUDE.md, AGENTS.md |
PreToolUse hook + rule files |
| Cursor | cursor-install |
.cursor/rules/superbrain.mdc |
Always-on rule file |
| Aider | aider-install |
.aider.conf.yml, AGENTS.md |
MCP server config + rules |
| Codex | codex-install |
AGENTS.md |
Rule file |
| OpenCode | opencode-install |
AGENTS.md |
Rule file |
| GitHub Copilot CLI | copilot-install |
~/.copilot/skills/superbrain/SKILL.md |
Global skill |
| VS Code Copilot Chat | vscode-install |
.github/copilot-instructions.md |
Repo instructions |
| Gemini CLI | gemini-install |
.gemini/settings.json, AGENTS.md |
BeforeTool hook + rules |
| Hermes | hermes-install |
~/.hermes/skills/superbrain/SKILL.md, AGENTS.md |
Global skill + rules |
| Kiro | kiro-install |
.kiro/skills/superbrain/SKILL.md, .kiro/steering/superbrain.md |
Skill + steering |
| Google Antigravity | antigravity-install |
.agent/rules/superbrain.md, .agent/workflows/superbrain.md |
Rules + workflows |
| OpenClaw | openclaw-install |
AGENTS.md |
Rule file |
| Factory Droid | droid-install |
AGENTS.md |
Rule file |
| Trae / Trae CN | trae-install |
AGENTS.md |
Rule file |
All install commands are idempotent — run them twice and nothing breaks.
Uninstall
Dedicated uninstallers exist for the three tools whose integration changes shared settings files:
bash
agsuperbrain claude-uninstall # removes hooks from .claude/settings.json
agsuperbrain cursor-uninstall # removes .cursor/rules/superbrain.mdc
agsuperbrain aider-uninstall # unpatches .aider.conf.yml
For the other platforms, installation is additive and non-destructive — just delete the file(s) listed in the matrix above.
MCP: what your assistant can do
When the MCP server runs (started automatically by the Aider, Claude Code, and Gemini integrations, or manually via agsuperbrain mcp-serve), these tools become available:
| Tool | Purpose |
|---|---|
search_code |
Semantic search across code and documents |
find_callers |
All functions that call a given function |
find_callees |
All functions a given function calls |
get_function_body |
Full source, docstring, and signature of a function |
path_between |
Call path from one function to another |
closure |
Transitive closure of any relationship |
get_subgraph |
Local neighborhood around a node |
stats |
Counts per node/edge type |
list_modules |
Every module in the graph |
list_functions |
Every function in the graph, optionally filtered by module |
These are called by your assistant, not by you — but if you're debugging, agsuperbrain mcp-serve lets you test them manually.
Using Super-Brain from agent frameworks
Super-Brain's MCP server works with any framework that speaks MCP:
- LangChain / LangGraph (via
langchain-mcp-adapters) - AutoGen (native MCP)
- CrewAI (MCP tool connector)
- SmolAgents
- Plain stdio JSON-RPC
LangChain / LangGraph
```python from langchain_openai import ChatOpenAI from langgraph.prebuilt import create_agent from langchain_mcp_adapters.tools import load_mcp_server
mcp = load_mcp_server(command="agsuperbrain", args=["mcp-serve"]) tools = mcp.get_tools()
llm = ChatOpenAI(model="gpt-4o") agent = create_agent(llm, tools, prompt="You are a codebase expert.")
response = agent.invoke({"messages": ["How does authentication work?"]}) ```
AutoGen
```python from autogen import ConversableAgent
agent = ConversableAgent( name="code_expert", llm_config={"model": "gpt-4o"}, tools=[{"command": "agsuperbrain", "args": ["mcp-serve"]}], ) ```
CrewAI
```python from crewai import Agent from crewai.tools import MCPTool
code_tools = MCPTool( server_name="agsuperbrain", server_command="agsuperbrain mcp-serve", ) expert = Agent(tools=[code_tools], role="Codebase Expert") ```
Manual MCP configuration (advanced)
If you prefer to wire the MCP server yourself instead of using an <ide>-install command, here are the configs:
Claude Code — ~/.claude/settings.json
json
{
"mcpServers": {
"superbrain": {
"command": "agsuperbrain",
"args": ["mcp-serve"]
}
}
}
Cursor — ~/.cursor/mcp.json
json
{
"mcpServers": {
"superbrain": {
"command": "agsuperbrain",
"args": ["mcp-serve"]
}
}
}
Note: cursor-install writes a rules file (.cursor/rules/superbrain.mdc) for always-on context. MCP is an additional, optional layer — set it up manually if you want both.
Aider — .aider.conf.yml
yaml
mcp-server: agsuperbrain:mcp-serve
Custom DB paths
Any MCP client can be configured to use a custom graph or vector path:
json
{
"mcpServers": {
"superbrain": {
"command": "agsuperbrain",
"args": [
"mcp-serve",
"--db", "/path/to/graph",
"--qdrant-path", "/path/to/qdrant"
]
}
}
}
Troubleshooting
The assistant doesn't see Super-Brain's tools
- Run
agsuperbrain doctorand confirm every component is green. - Confirm you've run
agsuperbrain ingest ./src— the graph must exist. - Restart your IDE — MCP servers are loaded at IDE startup.
- Check the IDE's MCP logs (varies by product). Claude Code logs to
~/.claude/logs/; Cursor logs underHelp → Toggle Developer Tools → Console.
"Connection refused" or "server failed to start"
- Make sure
agsuperbrainis on yourPATH(which agsuperbrain/where agsuperbrainshould print a path). - MCP uses stdio — nothing is listening on a network port, so firewall rules aren't the cause.
The assistant answers but without Super-Brain context
Re-run the install command for that platform. Integration files may have been overwritten by an update to the IDE or a linter.
Next steps
- CLI Reference — the commands behind every integration
- Architecture — what the MCP tools actually query
- Why Super-Brain — the problems this integration solves