M
MCP Directory CURATED DIRECTORY
911 MCP servers indexed
中文
← Back to directory
B

Blockbench MCP

Community
Adds an MCP server to Blockbench so AI clients can reach a local modeling session
How to install ↓ Compare GitHub source repository ↗
Category
Dev Tools #176 of 438
Stars
★ 596 Very popular
Transport
Streamable HTTP
Runtime
Node.js
Credentials
No credential needed
License
GPL-3.0
Last commit
56FMRS · C

Blockbench MCP delivers a local MCP server to Blockbench as a plugin, with README config examples for VS Code, Claude Desktop, Claude Code, Cline, Ollama, OpenCode, and Antigravity, under GPL-3.0. It suits users who want AI clients connected to a local modeling environment, but the README does not enumerate specific tools and it is not an official product.

Strongest · Setup experience 14/20 Weakest · Reliability 9/20

Reliability
9/20
Security and permissions
10/20
Maintenance
11/20
Documentation
12/20
Setup experience
14/20
Why each score
Reliability 9/20
The repository is presented as a plugin-distribution repo (README instructs loading mcp.js from jasonjgardner.github.io). No source files, tool inventory, CI workflow, or committed tests were supplied, so the MCP init + tool-list handshake cannot be verified and declared tools cannot be matched against real behavior. Under the static calibration rule, absence of verifiable execution evidence caps reliability at 12; only descriptive claims and a 438-star discovery signal (which adds nothing) are available, with error handling, dependency control, and edge cases unknown. Scored 9.
Security and permissions 10/20
No red-line behavior found: no credential theft, no covert exfiltration, and no real tokens or secrets in the install examples; the server binds only to local http://localhost:3000/bb-mcp. However, it is an unauthenticated local HTTP endpoint that any local process could connect to and drive Blockbench editing tools, and the plugin itself is loaded dynamically from a remote URL (jasonjgardner.github.io), raising supply-chain/update-trust concerns. No visible permission scoping, confirmation prompts for dangerous operations, or data-flow disclosure exist in the provided material. Risks are visible but mitigations are incomplete, so 10.
Maintenance 11/20
GPL-3.0 license is explicit, the repo is not archived, and there is only 1 open issue, suggesting an active project with clear ownership (jasonjgardner). But no release records, commit history, dependency-update practice, or security-response channel are provided, leaving governance and versioning gaps; the server path is the repo root and the manifest was not cached for inspection. Following the 'active but governance/versioning gaps' anchor, 11.
Documentation 12/20
Documentation is broad: plugin install URL, MCP Server port/endpoint settings, and client configs for General, VS Code, Claude Desktop (macOS/Linux and Windows variants), Claude Code, Antigravity, Cline, Ollama, and OpenCode, plus links to a sample project and Skills. Missing: tool parameter definitions, permission/security boundaries, cost or limits, and a troubleshooting section, and it does not disclose the implications of dynamically loading a remote plugin; some client configs (OpenCode, Antigravity) rely on screenshots instead of text, leaving hidden assumptions. 'Usable but with hidden assumptions and troubleshooting gaps' — 12.
Setup experience 14/20
The install path is clear: load the given URL in desktop Blockbench via File > Plugins > Load Plugin from URL, configure the port/endpoint, then fill in the mainstream client templates. Few steps, no build beyond the client runtime, cross-platform coverage, and a Windows-specific command. But it requires desktop (non-headless) Blockbench, and config correctness cannot be verified statically; there is no CI/test corroboration. Given the static cap of 15 and no execution evidence, 14.

Static review · not runListed 2026-09-18

Read the FMRS scoring method →

Fit and risk

Best for

  • 3D modelers using the desktop version of Blockbench
  • Developers who want MCP clients wired into their local modeling environment
  • Users adding AI assistance to a Blockbench workflow

Not for

  • Users who cannot run desktop Blockbench or load plugins
  • Clients without access to local HTTP endpoints
  • Users who need official support or a hosted service

Required permissions

  • Loading and running a third-party plugin inside Blockbench
  • Exposing a local MCP server port (default 3000)
  • AI client access to http://localhost:3000/bb-mcp

Risks and side effects

  • The plugin is third-party and not maintained by Blockbench itself
  • A local MCP endpoint without restrictions could be reached by other local programs
  • Incorrect port or endpoint configuration causes connection failures
  • Loading the plugin from a URL relies on remote script distribution

Setup

Before you start

Runtime：Node.js

  1. Open the desktop version of Blockbench, go to File > Plugins, click "Load Plugin from URL" and paste https://jasonjgardner.github.io/blockbench-mcp-plugin/mcp.js.
  2. In Blockbench, open Settings > General and configure MCP Server Port and MCP Server Endpoint (default :3000/bb-mcp).
  3. Configure the MCP server in your AI client, for example: npx mcp-add --type http --url "http://localhost:3000/bb-mcp" --scope project.
  4. For Claude Desktop, bridge with mcp-remote: npx -y mcp-remote http://localhost:3000/bb-mcp (on Windows, invoke through cmd /c).
  5. For VS Code use .vscode/mcp.json, and for Cline use cline_mcp_settings.json.
claude_desktop_config.json 
{
  "mcpServers": {
    "blockbench": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "http://localhost:3000/bb-mcp"]
    }
  }
}

Shown for Claude Desktop. Other clients may use a different file or key (VS Code uses "servers") — the configurator below converts it.

.vscode/mcp.json
{
  "servers": {
    "blockbench": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "http://localhost:3000/bb-mcp"
      ]
    }
  }
}

Goes in your project's .vscode/mcp.json (VS Code uses a "servers" key).

Terminal
claude mcp add blockbench -- npx -y mcp-remote http://localhost:3000/bb-mcp

Run it in a terminal; replace any <…> placeholders with your own values first.

Generate and validate client config

Check that it works

Confirm the plugin's server is running at the default http://localhost:3000/bb-mcp (Settings > General > MCP Server Port/Endpoint), then check that the client's tool list shows tools from the blockbench server.

Troubleshooting

  1. Confirm you are on desktop Blockbench and the plugin loaded successfully
  2. Check that MCP Server Port/Endpoint under Settings > General match the default :3000/bb-mcp
  3. Verify the client points at http://localhost:3000/bb-mcp
  4. On Windows, wrap npx with cmd /c for Claude Desktop
  5. Check that port 3000 is not in use by another program

Things to try

Once connected, you can ask your AI assistant things like:

  • List the MCP tools exposed by the Blockbench server
  • Inspect the model currently open in my Blockbench project
  • Run one of the sample-project prompts against the open Blockbench model

Use cases

Connect an AI client to a local Blockbench session over MCP
Use an AI assistant while modeling in Blockbench
Orchestrate tool usage through Agent Skills

Supported clients

VS Code
Claude Desktop
Claude Code
Cline
OllamaPartial support
OpenCodePartial support
Antigravity

Listed from the project's documentation, not tested by this site.

Overview

Blockbench MCP is a Blockbench plugin that runs an MCP server inside Blockbench, letting MCP-capable AI clients connect over HTTP to the local Blockbench instance. The plugin is installed by loading it from a URL, and the server port and endpoint are configured in Blockbench under Settings > General, with a default of :3000/bb-mcp. The README also mentions using Agent Skills to orchestrate tool usage and links to a sample project.

Similar servers

Context7 80 · B

Upstash's official server providing up-to-date third-party library docs for AI coding assistants

★ 64.1k · Tools 2 Compare with this →
Compare with other MCP servers

Source revision 97a8a80d3a37 Data synced 2026-11-02 Read the FMRS scoring method