Server Everything

What modelcontextprotocol/server-everything MCP server does

The modelcontextprotocol/server-everything MCP server is a reference and test implementation maintained in the official MCP servers repository. Its exposed tools operate on a knowledge graph made up of entities, relations, and observations. Agents can add records, inspect the graph, search for matching nodes, open named nodes, and remove selected data.

The repository describes Everything as a reference server covering MCP prompts, resources, and tools. The supplied tool list details the knowledge-graph portion: entities represent named items, relations connect those entities, and observations add information to existing entities. Relation text is expected to use active voice, which matters when an agent constructs graph statements.

Reach for this project when you need a compact target for MCP client testing, protocol demonstrations, or SDK-oriented development. It is less suitable when you need a production-ready memory store, because the repository explicitly positions its servers as educational reference implementations and asks developers to evaluate security requirements themselves.

Tools and capabilities

The available operations fall into four groups:

• Create graph data: create_entities adds multiple entities, create_relations connects multiple entities, and add_observations attaches observations to existing entities.
• Read graph data: read_graph returns the entire knowledge graph, while open_nodes retrieves specific nodes by name.
• Search graph data: search_nodes finds nodes using a query.
• Delete graph data: delete_entities removes entities and their associated relations, delete_observations removes selected observations, and delete_relations removes selected relations.

The batch-oriented names indicate that several entities, relations, or deletions can be submitted in one tool call. The material does not define the request schemas, matching rules, response formats, or storage backend, so those details should be checked in the source before building tightly coupled client logic.

How it works

An MCP client connects to the server and exposes its registered tools to an agent. The agent can then issue graph mutations or retrieval requests through MCP rather than calling an application-specific API directly. A typical exploration flow could create entities, connect them with active-voice relations, add observations, search for relevant nodes, and open or read the resulting graph.

The modelcontextprotocol/server-everything MCP server belongs to a repository of SDK reference implementations. Its purpose is to demonstrate MCP behavior and usage patterns, not to provide application-specific access controls, durable knowledge management, or a hardened deployment model.

Setup and configuration

The repository README states that TypeScript-based servers in the collection can be run with npx, and that servers should normally be configured inside an MCP client. It shows Claude Desktop configuration using a command and args entry, including the -y flag for non-interactive npx execution.

However, the supplied material does not identify the published package name or provide an exact npx command for the Everything server. Do not substitute the package names shown for the Memory, Filesystem, Git, or other example servers. Confirm the Everything package and its current launch instructions in the project source or release metadata before automating installation.

Limitations and notes

The modelcontextprotocol/server-everything MCP server is explicitly a reference implementation, not a production-ready solution. The repository advises developers to apply safeguards appropriate to their own security threat model. The provided information also does not establish whether graph contents persist between processes, whether authentication is supported, or how data is isolated between clients.

The repository uses Apache-2.0 for new contributions while existing code remains under MIT, according to its license note. Review the repository license files when redistributing or embedding specific code.