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

Unity API Documentation MCP Server

Community
Accurate Unity API documentation for AI agents, with no hallucinated signatures
How to install ↓ Compare GitHub source repository ↗
Category
Dev Tools #133 of 438
Stars
★ 67 Popular
Transport
stdio (local process)
Runtime
Python 3.10+
Credentials
No credential needed
License
MIT
Last commit
Tools
5
59FMRS · C

This is a third-party MCP server for Unity developers that serves accurate Unity API documentation from per-version SQLite databases, including method signatures, overloads, namespaces, and deprecation warnings. The README's benchmarks report 24/25 correct answers and zero hallucinated APIs for the MCP config, clearly ahead of Grep/Read-only configs. It supports Unity 6, 2023, and 2022 LTS and works with Claude Code, Cursor, and Windsurf without a Unity installation, with databases rebuilt weekly. Limitations: it does not index third-party assets, and the first run needs a network download.

Strongest · Documentation 14/20 Weakest · Reliability 9/20

Reliability
9/20
Security and permissions
11/20
Maintenance
12/20
Documentation
14/20
Setup experience
13/20
Why each score
Reliability 9/20
The static material contains only server.json, the README, and a project-structure listing; there is no verifiable execution evidence (no committed tests, no actual CI workflow content). The five declared tools (search_unity_api, get_method_signature, get_namespace, get_class_reference, get_deprecation_warnings) are consistent between manifest and README with no visible contradiction. On first run the server downloads a 20-30MB SQLite database from GitHub, an external dependency whose failure path is only vaguely covered by a troubleshooting table ("check your internet connection"), so error handling cannot be verified. The README claims weekly Monday rebuilds and startup freshness checks, but no workflow file corroborates this. Under static calibration, reliability must not exceed 12; absent tests and CI evidence, it is reduced to 9.
Security and permissions 11/20
No malware, credential theft, or covert exfiltration was found, so the 0-4 red-line band is not triggered. Static evidence is nonetheless limited: by default the tools download a database from GitHub into ~/.unity-api-mcp/ and write to the local filesystem, an unconfirmed write side effect, and the manifest declares no verification mechanism (hash or signature) for the network download, leaving supply-chain risk though no exploitation is evidenced. The UNITY_VERSION and UNITY_PROJECT_PATH variables are correctly marked non-secret, and UNITY_PROJECT_PATH only reads ProjectSettings/ProjectVersion.txt, a restrained scope. Main risks are visible while confirmation and integrity checking are incomplete, hence 11.
Maintenance 12/20
The MIT license is explicit, version 2.0.2 matches server.json, and the README shows a PyPI release badge, a weekly automatic Unity-release detection build, a db-v1 release page, and a sibling project (unreal-api-mcp), indicating sustained maintenance and clear ownership. However, this is a static review: actual commit history, issue response times, dependency-update policy, and security-response channel are not visible, and the 7 open issues cannot be assessed for responsiveness. Governance and versioning information has gaps, hence 12.
Documentation 14/20
Documentation is well layered: quick start, a version-detection priority table, the five tools with purposes and call examples, coverage plus explicit non-coverage (DOTween, VContainer, Newtonsoft.Json), a troubleshooting table, environment-variable table, multi-platform local database build commands, and benchmark methodology with a re-runnable script. A notable strength is disclosure of limits and cost (20-30MB first download, per-version databases). Deductions: benchmark charts are external image references that cannot be verified and the "Measured, not promised" claim is self-reported evidence; tool parameters lack full schema (input/output formats, error returns); and the README contains instruction-directed content aimed at AI agents, which under the evidence rules is not treated as trustworthy. Overall 14.
Setup experience 13/20
The installation path is clear and short: a one-line uvx command plus MCP config JSON, with client config snippets for mainstream clients (Claude Code, Cursor, Windsurf), plus alternative pip install and UNITY_PROJECT_PATH auto-detection, and an AI-assisted setup prompt. Windows and macOS are mentioned. Deductions: the first run requires an internet download of the database, and offline use needs a self-built database with extra ingest dependencies, a relatively fragile step; the Python 3.10+ prerequisite appears only in troubleshooting rather than before quick start; and actual client handshake connectivity is unverified. Under the static cap of 15, this scores 13.

Static review · not runListed 2026-09-10

Read the FMRS scoring method →

Fit and risk

What it can accessUses the networkConnects to a database

Best for

  • Teams developing Unity projects with Claude Code, Cursor, or Windsurf
  • Developers who want to avoid AI-invented Unity API signatures
  • Projects on Unity 6, Unity 2023, or Unity 2022 LTS
  • AI-assisted workflows that want to reduce source-file reads, saving context and tokens

Not for

  • Users who need third-party assets like DOTween, VContainer, or Newtonsoft.Json (not indexed)
  • Developers using Unreal Engine or non-Unity engines
  • Environments with Python below 3.10
  • Anyone needing official Unity support (this is an independent third-party project)

Required permissions

  • Read access to local MCP configuration files
  • Network access on first run to download the database (~20-30 MB)
  • Read/write access to ~/.unity-api-mcp/ to cache databases
  • Optional UNITY_PROJECT_PATH reads ProjectSettings/ProjectVersion.txt inside the project
  • Optional UNITY_INSTALL_PATH is used only for local database building (ingest) to locate the Unity install

Risks and side effects

  • First run or database updates require internet access; without it you must build the database locally
  • The downloaded database is ~20-30 MB and occupies local disk space
  • An incorrect UNITY_VERSION can cause API docs for the wrong version to be served
  • Databases are rebuilt weekly by CI, so version coverage may lag the newest Unity release
  • Third-party packages are not indexed and return no results

Setup

Before you start

Runtime：Python 3.10+

Other optional settings (3)
UNITY_VERSION optional Unity version to serve (e.g. 6000.3, 6, 2023, 2022); defaults to auto-detect or 6.
UNITY_PROJECT_PATH optional Path to a Unity project so the server can auto-detect the version from ProjectSettings/ProjectVersion.txt; not required.
UNITY_INSTALL_PATH optional Override Unity install path, used only for building databases locally (ingest).
  1. Add the server to your MCP config file (.mcp.json, mcp.json, or your tool's MCP settings): run unity-api-mcp with uvx and set UNITY_VERSION to match your project (e.g. "6000.3", "6", "2023", or "2022").
  2. Alternatively, install with pip install unity-api-mcp and set command to unity-api-mcp in the config.
  3. Instead of UNITY_VERSION, you can set UNITY_PROJECT_PATH to your Unity project so the server auto-detects the version from ProjectSettings/ProjectVersion.txt; if neither is set, it defaults to "6".
  4. On first run the server downloads the correct database (~20-30 MB) to ~/.unity-api-mcp/.
  5. Add the usage snippet from the README to your project's CLAUDE.md or equivalent instructions file so the AI knows when to reach for these tools.
.mcp.json 
{
  "mcpServers": {
    "unity-api": {
      "command": "uvx",
      "args": ["unity-api-mcp"],
      "env": {
        "UNITY_VERSION": "6000.3"
      }
    }
  }
}

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

.vscode/mcp.json
{
  "servers": {
    "unity-api": {
      "command": "uvx",
      "args": [
        "unity-api-mcp"
      ],
      "env": {
        "UNITY_VERSION": "6000.3"
      }
    }
  }
}

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

Terminal
claude mcp add unity-api -e UNITY_VERSION=6000.3 -- uvx unity-api-mcp

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

Generate and validate client config

Check that it works

After connecting, five unity-api tools (search_unity_api, etc.) should appear in the client's tool list; calling get_namespace("SceneManager") and getting "using UnityEngine.SceneManagement;" confirms the install works.

Troubleshooting

  1. If you see "Could not download Unity X database", check your internet connection or build locally with python -m unity_api_mcp.ingest --unity-version 2022
  2. If the wrong API version is served, set UNITY_VERSION explicitly and check the stderr line "unity-api-mcp: serving Unity <version> API docs"
  3. If the server won't start, check python --version (needs 3.10+) and verify the path with which unity-api-mcp or where unity-api-mcp
  4. If third-party packages return no results, that is expected: DOTween, VContainer, and Newtonsoft.Json are not indexed
  5. If the AI isn't using the tools, check that you added the CLAUDE.md snippet from the README

Things to try

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

  • Use get_method_signature to show all overloads of UnityEngine.Physics.Raycast
  • Which using directive do I need for SceneManager? Verify with get_namespace
  • Check whether the WWW API is deprecated and what replaces it
  • Search for APIs related to async scene loading with search_unity_api

Tools 5

search_unity_api read-only
Find APIs by keyword, e.g. "Tilemap SetTile" or "async load scene"
get_method_signature read-only
Get exact method signatures with all overloads, e.g. UnityEngine.Physics.Raycast
get_namespace read-only
Resolve using directives, e.g. "SceneManager" maps to using UnityEngine.SceneManagement;
get_class_reference read-only
Get a full class reference card, e.g. all methods, fields, and properties of "InputAction"
get_deprecation_warnings read-only
Check whether an API is obsolete, e.g. "WWW" should use UnityWebRequest instead

Use cases

Verify method signatures before writing Unity API calls instead of guessing
Look up the namespace for a type to add the correct using directive
See all members (methods, fields, properties) on a class
Check whether a Unity API is deprecated and what replaces it
Let AI agents generate compilable code in Unity projects

Supported clients

Claude Code
Cursor
Windsurf

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

Overview

This is an MCP server that gives AI agents accurate Unity API documentation, preventing hallucinated signatures, wrong namespaces, and deprecated API usage. It supports Unity 6 (one database per minor stream), Unity 2023, and Unity 2022 LTS, and works with Claude Code, Cursor, Windsurf, or any MCP-compatible AI tool. No Unity installation is required. On first run the server downloads the correct SQLite database (~20-30 MB) to ~/.unity-api-mcp/; all tool calls query the version-specific database and every query returns in under 15ms. Databases are rebuilt weekly so new Unity releases are covered automatically. The project does not cover third-party assets such as DOTween, VContainer, or Newtonsoft.Json.

Similar servers

Context7 80 · B

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

★ 63.3k · Tools 2 Compare with this →
Compare with other MCP servers

Source revision 3f65d1d46b20 Data synced 2026-10-19 Read the FMRS scoring method