Sandbox
@ashwwwin/furi

CLI and HTTP API for MCP server management

Furi manages MCP servers from the command line and through an HTTP API. It installs MCPs from GitHub, starts and stops them, shows logs and status, and can aggregate tools from running MCPs for client connections over SSE or stdio.

171 stars10 forksTypeScriptUpdated 1y ago
Who it's for

Builders who want to manage MCP servers and connect them to their agent client from one tool.

What it delivers

You can install, run, inspect, and expose MCP servers without juggling separate scripts and processes.

What it does

Install MCPs from GitHub

Adds public GitHub repos as MCP servers with `furi add <author/repo>`.

Start, stop, and restart servers

Controls individual MCP processes with `furi start`, `furi stop`, and `furi restart`.

List tools and call them

Shows available tools for a running MCP with `furi tools <mcpName>` and calls them with `furi call <mcpName> <toolName> ...`.

Show status and logs

Lists installed and running MCPs, and can show per-server status and log lines with `furi status`.

MCP aggregator

Combines tools from all running MCPs and exposes them through SSE with `furi meta start` or stdio with `furi connect`.

HTTP API proxy

Starts HTTP routes for running MCPs with `furi http start`, including public tool routes and optional sudo management routes.

Config and process storage

Stores installed MCPs, config, and logs in `.furikake` and uses PM2 for process state management.

How to get it

  1. 1To install Furi, you can use the following command
    curl -fsSL https://furi.so/install | bash
  2. 2Verify the installation by running
    furi
  3. 3To upgrade Furi to the latest version, run
    furi upgrade
  4. 4Furikake works with any public github repo as follows
    furi add <author/repo>
  5. 5You can also rename an MCP by using the rename command, please note this will restart…
    furi rename <old-name> <new-name>
  6. 6Run
    furi remove <mcpName>

README

furi

furi is an easy to use, CLI & API for MCP management.

  • Download MCP servers [from GitHub]
  • Smithery.yaml detection (or auto detects/handles execution)
  • Fully featured CLI [nanospinners, readability]
  • Typescript & Javascript MCP's are supported
  • Python based MCP's are a key roadmap item (and will be supported)
  • HTTP API Routes (uses Bun http, stdio to http, clear and standard routes)
  • Customizable port and visibility of sudo routes
  • View all running MCPs + logs for each process
  • Process state management with PM2
  • Built with Bun and Typescript
  • is good with rice

Installation (macOS/Linux)

To install Furi, you can use the following command:

curl -fsSL https://furi.so/install | bash

Verify the installation by running:

furi

Furikake uses Bun under the hood, the install script will install Bun if it is not already installed.

Upgrade Furi

To upgrade Furi to the latest version, run:

furi upgrade

How to use

Manage MCPs

Furikake works with any public github repo as follows:

furi add <author/repo>

eg. furi add smithery-ai/mcp-fetch

You can also rename an MCP by using the rename command, please note this will restart the MCP if it is running.

furi rename <old-name> <new-name>

eg. furi rename smithery-ai/mcp-fetch mcp-fetch

Delete an MCP

furi remove <mcpName>

eg. furi remove mcp-fetch

List installed MCPs

Show all installed MCPs

furi list

Start an MCP

furi start <mcpName> -e '{"name1":"value1", "name2":"value2"}'

-e env is optional and dependant on the MCP server being called

Ensure you pass a valid JSON object to the -e flag.

Once you start a server with the -e flag, it will be saved to the config file and re-used when using the server again.

In order to view the env variables required for an MCP, use:

furi env <mcpName>

You can get a list of all the tools available (with details) of any MCP by using:

furi tools <mcpName>

then you can call the tool with:

Call a tool

furi call <mcpName> <toolName> '{"param1":"value1", "param2":"value2"}'

Parameters must be a valid JSON string enclosed in single quotes

Stop an MCP

furi stop <mcpName>

Restart an MCP

furi restart <mcpName>

Get the status of all running MCPs

This will show you the status of all running MCPs.

furi status

If you want to get the logs a specific MCP, you can use:

furi status <mcpName>

to view more output lines, use -l <lines>

Configuration storage

All installed MCPs, your configuration and logs are stored in the .furikake directory which can be located by running:

furi where

Using the MCP Aggregator

You can use Furikake with any MCP client such as Cursor via the MCP Aggregator.

Furi collects tools from all running MCPs and exposes them through an SSE endpoint that your app or mcp client can subscribe to. The aggregator automatically builds the list of tools from all running MCPs and listens for new tools as MCPs are started and stopped.

For MCP Clients that support SSE

To start the aggregator server:

furi meta start

This will also show you the endpoint your MCP client needs to subscribe to

For MCP Clients that only support stdio

Some MCP clients don't support SSE transport and require stdio connections. For these clients, use:

furi connect

This starts the aggregator server directly in stdio mode, allowing MCP clients to connect via stdin/stdout. The server will aggregate all tools from your running MCPs and make them available through the stdio transport.

Important: This command provides JSON-only output to comply with MCP protocol requirements. All logging is suppressed to ensure clean communication with MCP clients.

Note: Unlike furi meta start, this command runs in the foreground and maintains the connection directly.

You can specify a custom port:

furi meta start -p 9338

If you don't pass a port, it will default to 9338

To stop the aggregator:

furi meta stop

To restart the aggregator (preserving port settings):

furi meta restart

To check the status of the aggregator server:

furi meta status

To view more output lines, use -l <lines>

Using the HTTP API

  • Any MCP that is running, will automatically have an http route.
  • Turning an MCP on/off can only be done via the cli.

To access your MCP's via http, you can turn on the proxy via:

furi http start

In order to pass a port, you can use the http start -p <port> flag.

furi http start -p 9339

If you don't pass a port, it will default to 9339

To turn off the route, you can use:

furi http stop

HTTP API Reference

The Furikake HTTP API is divided into public routes and sudo routes. Public routes are accessible by default, while sudo routes must be explicitly enabled. With sudo routes, you can actively manage packages and instances via the HTTP API.

If you want to secure your API, you can set the HTTP_AUTH_TOKEN environment variable.

export HTTP_AUTH_TOKEN=your-secret-token

or in your .env file:

HTTP_AUTH_TOKEN=your-secret-token

GitHub Rate Limiting

To avoid GitHub API rate limiting when installing packages, you can provide a GitHub personal access token:

export GITHUB_KEY=your-github-token

or in your .env file:

GITHUB_KEY=your-github-token

This increases your rate limit from 60 requests/hour to 5,000 requests/hour. You can create a personal access token at GitHub Settings > Developer settings > Personal access tokens.

API Response Format

All API endpoints follow a standardized JSON response format:

  • Success responses:

    {
      "success": true,
      "data": {"The response varies by endpoint"}
    }
    
  • Error responses:

    {
      "success": false,
      "message": "Descriptive error message"
    }
    

HTTP Methods

  • POST - Used only for /mcpName/call/toolName and /mcpName/start endpoints
  • GET - Used for all other endpoints

Public Routes

EndpointMethodDescriptionParametersResponse Format
/listGETList running MCPs?all=true (optional) to show all installed MCPs{"success": true, "data": ["mcpName1", "mcpName2"]}
/toolsGETList all available tools from all running MCPsNone{"success": true, "data": [{"name": "toolName", "description": "Tool description", "inputSchema": {...}, "mcpName": "mcpName"}]}
/<mcpName>/toolsGETList tools for a specific MCPNone{"success": true, "data": [{"name": "toolName", "description": "Tool description", "inputSchema": {...}}]}
/<mcpName>/call/<toolName>POSTCall a tool on an MCPTool parameters as JSON in request body{"success": true, "data": {/* Tool-specific response */}}

Example Usage:

List running MCPs:

curl http://localhost:9339/list

To view all available tools for all online MCPs, you can use:

curl "http://localhost:9339/list"

List tools for all online MCPs:

curl http://localhost:9339/tools

List tools for a specific MCP:

curl http://localhost:9339/<mcpName>/tools

Call a tool:

curl -X POST http://localhost:9339/<mcpName>/call/<toolName> -d '{"data1":"value1", "data2":"value2"}'

Sudo Routes

To enable sudo routes that allow API management of MCPs:

furi http start --sudo
EndpointMethodDescriptionParametersResponse Format
/statusGETGet status of all MCPs (running and stopped)None{"success": true, "data": [{"name": "mcpName", "pid": "12345", "status": "online", "cpu": "0%", "memory": "10MB", "uptime": "2h"}]}
/add/<author>/<repo>GETInstall MCP from GitHubNone{"success": true, "data": {"installed": true, "message": "Successfully installed"}}
/<mcpName>/statusGETGet status of a specific MCP?lines=10 (optional) to control log lines{"success": true, "data": {"name": "mcpName", "pid": "12345", "status": "online", "logs": ["log line 1", "log line 2"]}}
/<mcpName>/restartGETRestart a specific MCPNone{"success": true, "data": {"restarted": true}}
/<mcpName>/startPOSTStart a specific MCPEnvironment variables as JSON in request body{"success": true, "data": {"started": true}}
/<mcpName>/stopGETStop a specific MCPNone{"success": true, "data": {"stopped": true}}
/<mcpName>/renameGETRename a specific MCP?newName=<newName> (required){"success": true, "message": "Renamed from oldName to newName"} or {"success": false, "message": "Error message"}
/<mcpName>/removeGETDelete a specific MCPNone{"success": true, "data": {"removed": true}}
/<mcpName>/envGETGet environment variables for a specific MCPNone{"success": true, "data": {"variables": ["key1", "key2"]}}

Example Usage:

Get status of all MCPs:

curl http://localhost:9339/status

Install an MCP:

curl http://localhost:9339/add/<author>/<repo>

Get status and logs of a specific MCP:

curl "http://localhost:9339/<mcpName>/status?lines=20"

Start an MCP with environment variables:

curl -X POST http://localhost:9339/<mcpName>/start -d '{"API_KEY":"your-api-key-here"}'

Restart an MCP:

curl http://localhost:9339/<mcpName>/restart

Stop an MCP:

curl http://localhost:9339/<mcpName>/stop

Rename an MCP:

curl "http://localhost:9339/<mcpName>/rename?newName=<newName>"

Remove an MCP:

curl http://localhost:9339/<mcpName>/remove

If you face any issues with the HTTP API server, you can use the furi http status to debug.

Closing notes

If you've made it this far, I hope you find Furikake useful and time saving. I built this for fun as a way for me to work with MCP's more hands on. If you wish to contribute, feel free to open an issue or a pull request. I will merge after I check out your changes.

If you think this is a good idea, please star the repo.

Thanks for checking out Furikake.

Files in the repo

Repository payload16 top-level entries
  • .github
  • app
  • tests
  • .cleanup
  • .dockerignore
  • .gitattributes
  • .gitignore
  • bun.lock
  • Dockerfile
  • fly.toml
  • index.ts
  • install.sh
  • LICENSE
  • package.json
  • README.md
  • tsconfig.json

Discussion (0)

Ask about usage, or say what you built with it

Sign in to join the discussion.

No comments yet. Be the first to say what this is good for.

More tools

JuliusBrussee/
caveman

🪨 why use many token when few token do trick — Claude Code skill that cuts 65% of tokens by talking like caveman

105k
1 add
MemPalace/
mempalace

The best-benchmarked open-source AI memory system. And it's free.

59k
stablyai/
orca

Orca is the ADE for working with a fleet of parallel agents. Run any coding agent with your own subscription. Available on desktop, mobile and remote runtime.

66k

A cross-platform desktop All-in-One assistant for Claude Code, Codex, OpenCode, OpenClaw, Grok Build & Hermes Agent. Only official website: ccswitch.io

132k

Never stop coding. Free MIT AI gateway: one endpoint, 352 providers (150+ free), 1200+ models Kimi, Claude, GPT, Gemini, GLM, DeepSeek, MiniMax. Works with Claude Code, Codex, Cursor, OpenCode, Cline & Copilot. Quota-aware auto-fallback, RTK+Caveman compression saves 15-95% tokens, MCP/A2A, Desktop/PWA. Built by 550+ contributors

64k
headroomlabs-ai/
headroom

Compress tool outputs, logs, files, and RAG chunks before they reach the LLM. 20% fewer tokens for coding agents, 60-95% fewer tokens for JSON, same answers. Library, proxy, MCP server.

71k