05. Resources in MCP

Chapter 5 of 22 · 20 min

KEY INSIGHT

Resources expose data to AI applications through a URI-based system with optional change subscriptions that keep AI context current. Resources in MCP represent data that servers expose to clients. Unlike tools, resources do not perform actionsΓÇöthey provide information. AI applications request resources when context is needed for task completion. **Resource Structure:** Each resource has a URI, name, description, and MIME type. URIs follow a custom scheme that the server defines. ```python from mcp.types import Resource, ResourceTemplate # Single resource file_resource = Resource( uri="file:///config/app.yaml", name="application_config", description="Current application configuration", mimeType="application/yaml" ) # Resource template (dynamic resources) config_template = ResourceTemplate( uriTemplate="file:///config/{filename}", name="dynamic_config", description="Configuration files by name", mimeType="application/yaml" ) ``` **Implementing Resource Handlers:** Servers must implement handlers for listing and reading resources. ```python @app.list_resources() async def list_resources() -> list[Resource]: return [ Resource( uri="file:///data/status.txt", name="status", description="System status file", mimeType="text/plain" ) ] @app.read_resource() async def read_resource(uri: str) -> str: if uri == "file:///data/status.txt": with open("/data/status.txt", "r") as f: return f.read() raise ValueError(f"Unknown resource: {uri}") ``` **Resource Templates:** Templates handle dynamic resource URIs. The client provides substituted values when requesting. ```python @app.list_resource_templates() async def list_resource_templates() -> list[ResourceTemplate]: return [ ResourceTemplate( uriTemplate="file:///users/{user_id}/profile", name="user_profile", description="User profile by ID" ) ] @app.read_resource() async def read_resource(uri: str) -> str: # Parse template variables from URI if uri.startswith("file:///users/"): user_id = uri.split("/")[3] return get_user_profile(user_id) ``` **Change Notifications:** Resources can notify clients when content changes. This keeps AI context accurate without repeated polling. ```python # Server sends notification when resource changes await app.request_context.session.send_resource_updated( uri="file:///data/status.txt" ) ``` Clients subscribe to resources during listing. Servers must track subscriptions and send updates accordingly. **Failure Modes:** - Missing resource files return appropriate errors, not empty strings - Large resources should stream or limit size - Permissions errors surface as specific error codes, not generic failures

Local verification checkpoint

Run the smallest example from this chapter in a local workspace and record the package version, runtime, data path, and observed output. If the result depends on model size, vector count, CPU/GPU backend, or available memory, note that constraint beside the exercise so the lesson remains reproducible.

EXERCISE

Build a resource server that exposes two resources: a static "about" document and a dynamic resource showing the server's current memory usage. Implement subscriptions for the memory resource.