Jinny in a modern office with the local SindByte MCP workspace

SindByte MCP Server

Connect local AI to real Windows work: manage files, operate applications, run timers and call controlled tools directly from LM Studio, Roo Code or another MCP host.

What it is
Local MCP server for Windows: it gives AI hosts such as LM Studio or Roo Code tools for files, desktop and timers.
Who it is for
For anyone who wants to connect a local AI model to real Windows tasks, privately or in a team.
Requirements
Windows 10 or 11, 4GB RAM minimum, about 50MB disk space, and an MCP-capable AI client (like LM Studio). Internet only for web-based tools and external APIs. FAQ
How to get it
As a ZIP package via the download button above. Older versions are listed under “Old downloads”.
Cost
Private, non-commercial use is free. Business, team, and commercial usage goes through Smart Package Robot or by email to info@it-berater.org.

Your advantage: one local endpoint brings the tools together while approvals, filters and credentials determine exactly what each host may use.

Release 9.1.53: the current package supports IPv6 and IPv4 for SFTP, tries alternative server addresses within a shared connection deadline, and closes failed connection sockets cleanly. This helps when a hosting provider is reachable over IPv6 but its IPv4 endpoint times out.

Quick start: SindByte runs locally at http://127.0.0.1:5555/mcp by default. The Config Editor manages the port, approvals and published tool groups in one place.
Jinny with SindByte local tools and Windows workstation
Jinny with the local tool surface used by SindByte MCP Server.

Need help with setup and operation?

For companies we handle installation, model selection matched to your hardware, safe permissions and team training on request. Start with a free 20-minute intro call.

239Host-callable tools
19MCP families
5555Default port
LOCALData stays under your control

Live Runtime Surfaces

See the product before you install it. These images show surfaces of the Windows application (screenshots, some from older versions) and one illustration.

Illustration: magic lamp as a symbol for the SindByte MCP server
Illustration. You see the endpoint and runtime state in the main window of SindByte 01.exe.
Config Editor with annotations
Config Editor for feature flags, provider keys, port selection, safety posture, and publication behavior.
Dialog-LAB live interface
Dialog-LAB is the controlled review surface for multi-model rounds and reasoning extraction.
SindByte main window (version 1.8.69) with the Dialog-LAB and Chat-Window buttons
LMChat is the lighter local chat surface when you want faster one-model execution with IQ support.
Timer interface
Timer workflows turn prompt logic into scheduled local automation and repeatable review loops.
Trading Hub live interface
Trading Hub is a credential-aware operator surface that sits beside the core MCP runtime.

Runtime Architecture

1. Host

LM Studio, Roo Code, Cline-style agents, VS Code MCP add-ons, or custom clients open the MCP session.

2. Endpoint

SindByte exposes streamable HTTP JSON-RPC at /mcp on port 5555 by default.

3. Publication Layer

Feature flags, category filters, credentials, and registration mode decide what the host sees.

4. Execution Layer

Files, windows, timers, reasoning helpers, images, and credential-backed routes remain under local operator control.

MCP detail: custom clients can work against the standard lifecycle with initialize, tools/list, and tools/call. Many hosts cache schemas, so reload the host after changing filters, credentials, or registration mode.

How SindByte fits into your workplace

Host ecosystem around SindByte
Connect several MCP-capable hosts to the same local runtime without building a separate tool environment for every client.
Runtime surface graphic
Configuration, chat, timers and optional specialist surfaces work around the same local endpoint.
Config surface control graphic
Feature flags, credentials, filters, and publication mode shape the live catalog. This visual is the shortest explanation for why tool counts vary.

Deployment Patterns

Jinny connects several AI hosts to one local MCP hub
One local SindByte endpoint can connect several compatible AI hosts to the same controlled tool surface.
Host Best fit What it gets from SindByte Operational note
LM Studio Local private workflows and model experiments Local inference plus local file, desktop, timer, and reasoning routes on the same Windows machine Good first deployment because the model host and tool runtime stay on one PC
Roo Code / Cline Coding, repo maintenance, ops workflows FileTools, DataTools, TextTools, CUTools, SystemTools, IQ helpers, and image routes Use ask-first on destructive routes and reload after schema changes
VS Code MCP add-ons and agent CLIs Structured agent runs and bounded automation Consistent MCP endpoint for diffs, file actions, screenshots, and tool-assisted review loops Short registration is useful when a host prefers a tighter catalog
Custom MCP clients Internal apps, dashboards, orchestration layers Standard MCP JSON-RPC plus a runtime surface that can be intentionally narrow or intentionally broad Live tool count depends on credentials, filters, and registration mode by design

Startup Examples

PS C:\SindByte> .\SindByte 01.exe Main shell visible Listening on port 5555 Endpoint: http://127.0.0.1:5555/mcp
Practical check: start with spr_server_assistant or ToolCatalog.ListCategories when you want to inspect the current runtime surface before deeper automation.

LM Studio

"mcpServers": { "sindbyte-mcp": { "url": "http://127.0.0.1:5555/mcp" } }
Use the local MCP URL and reload LM Studio after config changes so it picks up the current catalog.

Roo Code / Streamable HTTP

"mcpServers": { "sindbyte-mcp": { "type": "streamable-http", "url": "http://127.0.0.1:5555/mcp" } }
Use the explicit transport in hosts that distinguish streamable HTTP from other MCP transports.

Registration, Filters, and Safety

Jinny protects local files and approved MCP tools
Local data, credentials and tool approvals stay under your control through filters and confirmation rules.
Registration mode graphic
Registration mode changes how the runtime is published to the client, not what the whole product is capable of.

Full Registration

Enabled routes are published more directly so the host can see the broadest practical tool list without a category bridge.

Short Registration

Hosts inspect categories and tool metadata through ToolCatalog.ListCategories, ToolCatalog.ListTools, ToolCatalog.GetToolSchema, and ToolCatalog.CallTool.

Category Filters

ON publishes enabled tools, ESSENTIAL publishes only essential routes, and OFF keeps the category out of the live catalog.

Safety States

Individual tools can be Disabled, Ask First, or Enabled. Use ask-first on deletes, execution, registry changes, and trading actions.

Short-mode detail: include_schema=true is useful when a host wants richer tool definitions. This matters most when a client is navigating the catalog instead of receiving a broad direct publication.
Config Editor technical screenshot
The Config Editor is not a separate shadow system. It writes the exact runtime configuration the server uses on startup and refresh.

What the Config Surface Controls

  • Feature flags and essential bits for the audited default product surface.
  • Provider credentials for OpenAI, OpenRouter, Serper, trading venues, and enterprise-style connectors before those routes are expected in a client.
  • Publication behavior such as registration mode, category filters, and safety posture.
  • Runtime settings including port, timeouts, log level, image model selection, and local-only behavior.
  • Profiles and batch actions that make minimal, coding, image, desktop, or trading-oriented deployments practical without hand-editing JSON.

Why This Positioning Matters

Local Automation Layer

Files, clipboard, windows, system state, screenshots, and data processing are handled where the work actually lives: on the Windows machine.

Reasoning Surfaces

Dialog-LAB, LMChat, IQ helpers, and timers make SindByte a guided operator environment rather than a raw transport shim.

Credential-aware Connectors

Trading, cloud image, email, and enterprise connectors appear when credentials and filters allow them, which keeps each runtime intentionally scoped.

Standard Host Compatibility

The same runtime can serve a local model desktop, a coding assistant, or a custom internal client without changing the core architecture.

Next Steps

Jinny opens the practical MCP toolbox
From download to the first file, desktop or timer workflow, Jinny guides the next steps.

Download

Get the portable package and start the runtime on your Windows machine.

Download

Connect a Host

Use client snippets for LM Studio, Roo Code, and custom MCP patterns.

Open Connections

Manual

Read setup, provider, filter, and troubleshooting guidance in one place.

Read Manual

Tool Map

Browse families and understand which routes are local, filtered, or credential-gated.

View Tools

Explore SindByte

Open a focused guide instead of scrolling through one long product page.

Download Activity

Daily activity for every offered download over the rolling last 28 days. Release markers identify new product versions.