What Is an MCP Server and Why Your Game Asset Tool Needs One

Assistants are learning to use tools

Claude and ChatGPT began as text generators. They learned to browse the web, run code and read files. Now they are learning the software your team already runs: project trackers, databases, design systems, asset catalogs.

The Model Context Protocol (MCP) is the standard that carries them there. Anthropic developed it and the wider ecosystem adopted it, and it defines how an assistant connects to an external tool, discovers what that tool can do, and calls it with proper authentication. One interface, so any compatible assistant plugs into any compatible tool.

How MCP works in practice

An MCP server publishes a set of tools, each a discrete operation with a name, a description and a schema for its inputs and outputs. On connecting, an assistant:

1. Discovers the tools, reading the list of operations and their descriptions
2. Picks one, matching those descriptions against what you asked for
3. Calls it with structured parameters and reads a structured response
4. Chains several when the task needs more than one

The assistant owns the language and the formatting. The server owns data access, validation and business rules. Neither has to know how the other is built.

What PolyDrobe's MCP server exposes

PolyDrobe runs an MCP server at https://polydrobe.app/mcp with 51 tools across the catalog:

• Projects: list, read, create, update and delete projects
• Assets and variants: full CRUD on both, plus a flat variant listing per project
• Classification: categories, tags, rarities, releases, currencies and statuses
• Tags: add and remove tags on assets and variants
• Team: list members, change roles, remove people
• Search: full-text search across assets and variants
• Activity: read the project activity log
• Images: fetch a variant thumbnail, image or texture, and set a thumbnail
• Bulk: create, update or delete many variants in one call
• Data: export a project, import a payload back

Every call is project-scoped and runs under the same role-based access control as the web interface. An Editor's token creates assets. A Viewer's token reads them.

What the tools do together

A single tool is a single call. The value shows up when an assistant chains them against a sentence you said out loud.

Ask questions about your catalog

"How many variants in Project Orion are still in Review status?"

The assistant calls list_projects to find Project Orion, calls list_assets with a status filter, adds up the variants and answers with the number and the context around it.

Create assets from descriptions

"Create an asset called Night Ops Vest in the Outerwear category with Epic rarity. Add three variants: Midnight Black, Forest Green and Arctic White, all priced at 1500 coins."

One create_asset call, three create_variant calls. A minute of clicking through forms collapses into one sentence.

Search and compare

"Find all Legendary-rarity helmets and list their prices."

The assistant runs search or a filtered list_assets, pulls the price off each variant, and lays them out as a table.

Generate reports

"Give me a release readiness summary for Release 2.1. How many variants are Approved against still In Progress?"

It queries the project, filters by release version, groups by status and writes the summary, in place of the filtering and counting somebody would otherwise do by hand.

How to connect

The server authenticates with OAuth 2.1. To connect an assistant:

1. Get a Bearer token from PolyDrobe's OAuth endpoint at /o/token/
2. Point the assistant at https://polydrobe.app/mcp
3. Ask it something, and it discovers the tools on its own

For Claude, add the server to your Claude configuration. For assistants that speak MCP or OpenAPI, the plugin manifest at /.well-known/ai-plugin.json and the OpenAPI spec at /api/openapi.json carry the connection details.

Why this matters for game teams

Asset management runs on repetitive queries and data entry. Asking about the status of the Season 3 battle pass items should not mean logging in, finding the project, setting two filters and scanning a list. It should be a question you ask while reading a design doc or sitting in a planning meeting.

MCP also opens the door to automation: a CI step that verifies asset completeness before a release build, a Slack bot that answers what shipped in the last update, a planning script that reads the catalog and reports the gaps.

An asset catalog your assistants can read is a catalog your team queries in the moment the question comes up, rather than the afternoon somebody sets aside for it.

Key takeaways

• MCP is the standard that lets an AI assistant connect to an external tool and call operations on it.
• PolyDrobe's MCP server exposes 51 tools covering projects, assets, variants, classification, team, search, media, bulk edits and data exchange.
• Assistants chain those tools into workflows: catalog queries, batch creation, release readiness reports.
• Authentication runs on OAuth 2.1 under the same role model as the web interface.
• The server is live at https://polydrobe.app/mcp, and any MCP-compatible assistant can connect to it.