⌕
Explore your codebase
Inspect files, understand project structure, search source code, and review implementation details in authorized workspaces.
Local-first AI development · MCP toolsYour AI. Your code.
Your AI. Your code.
Your workstation.
Connect compatible AI assistants to your development workspace with SFI Agent. Inspect code, make edits, run approved commands, and review Git changes—while keeping execution on your own machine.
Open source · Runs locally · Workspace-scoped access · You control the tunnel
Terminal — SFI Agent● local
# 01 · Install the host
$ curl -fsSL https://agent.shahfahadinc.com/install.sh | bash
# 02 · Start the control center
$ sfi-host --web
✓ Local control center ready
↳ http://localhost:5180
✓ MCP tools available
↳ Workspaces · Files · Git · Commands
! Start a tunnel only when remote access is needed.
Protected workspace accessLocal-first architecture
BUILT AROUND YOUR WORKFLOW⌘ Local workspaces↔ Model Context Protocol⌁ Cloudflare Tunnel⇄ ngrok⑂ Git-aware tools
A useful local bridgeGive your AI the tools.
Give your AI the tools.
Keep yourself in control.
SFI Agent connects an AI client to tools running on your computer. You decide which projects are available and which actions are permitted.
✎
Make focused changes
Create files and edit code through tools designed to respect workspace boundaries and protect sensitive paths.
⌘
Run your workflow
Use approved commands and tests, inspect Git status and diffs, and keep the final verification loop close to the code.
⌁
Choose your connection
Work locally, or use Cloudflare Tunnel or ngrok when an internet-accessible MCP endpoint is needed.
◈
See what is happening
Use the local control center to manage workspaces, inspect connection details, and monitor tool activity supported by your build.
⛨
Security-minded by design
Workspace allowlists and secret protections help reduce accidental exposure. Review permissions before enabling remote access.
Start in minutes
From install to your first tool call.
Install the host, open the local control center, then connect an MCP-compatible client using the endpoint and transport supported by your setup.
✓
One-command installer
Installs the host and its required components where supported.
✓
Visual control center
Manage workspaces and connection settings locally.
✓
Step-by-step client guides
Follow the guide for your specific AI client and account capabilities.
◉ Connection Hub◌ Live Activity▤ Workspaces✧ Skills & Tools
LOCAL SERVICE
Workspace connection
Select the project the AI can access.
▣ my-project Authorized
⌁ Tunnel Optional
Workspace scoped
Connect your stack
Use the client that fits your workflow.
Client support and remote MCP availability can vary by product, plan, operating system, and current client version. Follow the relevant setup guide.
✳
ChatGPT
Configure a remote MCP connection where custom MCP apps/connectors are available to your account.
ChatGPT setup →◉
Claude
Choose the appropriate local or remote MCP workflow supported by your Claude client.
Claude setup →⌘
Other MCP clients
Use the connection details and transport supported by your client, such as Cursor or compatible tools.
All integrations →Good to know
Frequently asked questions
Does SFI Agent upload my whole project to a cloud service?
SFI Agent is designed around local execution. When you connect a remote AI client, the client may send selected prompts and tool results to its own service. Review your AI provider's data policies and only authorize workspaces you trust.
Do I need Cloudflare Tunnel or ngrok?
No, not for every workflow. A local MCP client can use a local server if it supports that connection mode. A remote web client usually needs a reachable HTTPS endpoint; a tunnel such as Cloudflare Tunnel or ngrok can provide one without directly opening a router port.
Is the tunnel URL the same as my local control center?
Not necessarily. The local dashboard (shown here at localhost:5180) and the MCP server endpoint are separate routes or services in your configuration. Publish only the intended MCP service, not the entire local admin dashboard.
Does one install work with every AI client?
No single setup is guaranteed across all clients. MCP support, transport, account eligibility, authentication and local/remote connection requirements differ. Check the client-specific guide and test the endpoint before use.
Can it run terminal commands and edit files?
Available tools depend on your installed build and configured permission mode. Review command policies, workspace boundaries and Git changes before approving consequential actions.
Your next step
Make your local workflow AI-ready.
Follow the install guide, configure your first workspace, and choose a connection method that matches your client.