Definition
An MCP server registry is the collection of tool schemas (names, descriptions, parameter definitions) that an MCP-compatible agent loads at session startup, consuming context tokens proportional to the number and complexity of registered tools.
In Depth
When Claude Code, Cursor, or another MCP client starts a session, it loads tool definitions from every configured MCP server. Each tool contributes its name, description, and JSON Schema parameters to the agent's context window. A server with 11 tools (like Scavio's MCP endpoint) might add 800-1200 tokens to context. A server with 30+ tools can add 3000+ tokens. This creates a hidden cost: even before the agent does any work, its context is partially consumed by tool schemas. The practical impact is that agents with 5+ MCP servers enabled may spend 5-10% of their context budget on tool definitions alone. Mitigation strategies include: disabling unused MCP servers per session, using pre-tool-use filtering (Gandalf pattern) to limit which tools the agent considers, and preferring MCP servers with fewer, well-scoped tools over sprawling registries.
Example Usage
A developer notices Claude Code sessions are slower and more expensive than expected. They audit their MCP registry: 6 servers contributing 47 total tools, consuming ~4500 tokens before any work begins. They disable 3 rarely-used servers, reducing registry overhead to ~1800 tokens and saving roughly $0.02 per session on Opus-tier models.
Platforms
MCP Server Registry is relevant across the following platforms, all accessible through Scavio's unified API:
Related Terms
Model Context Protocol (MCP)
Model Context Protocol (MCP) is an open standard that defines how large language models discover and invoke external too...
Context Bloat
Context bloat is the accumulation of tokens in an LLM's context window before the user has asked anything — usually from...
Agent Token Budget
An agent token budget is a programmatic limit on how many context tokens an AI agent allocates to tool call results (par...