Model Context Protocol
Also known as MCP, Model Context Protocol server
Definition
Model Context Protocol, or MCP, is an open protocol for connecting AI applications with external tools, resources, and prompts through a standardized interface. It gives clients and servers a common way to describe and invoke capabilities.
What MCP standardizes
An MCP client, such as an AI application, can connect to an MCP server that exposes tools, resources, and prompts. The server describes what it offers, and the client can request a capability through the protocol. This shared interface can reduce one-off integration work across applications and data systems.
The protocol does not decide whether a tool call is correct, whether a model should be trusted, or whether a result is safe to use. Those decisions belong to the application, its permissions, and its review controls. A server may expose a powerful operation, so teams should apply least privilege and clear approval boundaries.
Observe the complete interaction
A useful record includes the task identifier, model and prompt context, selected tool, arguments, authorization result, response, timing, errors, retries, and final outcome. Avoid storing credentials or unnecessary sensitive payloads. Validate arguments and treat tool results as external input that may be incomplete or adversarial.
Evaluate connected workflows
Test whether the model selects the right tool, supplies valid arguments, handles failures, and uses the result correctly. Include permission failures and unavailable servers in the evaluation dataset. MCP can make integrations more portable, but good engineering analytics still needs to connect tool behavior with task success, latency, and cost.
How this relates to Weave
MCP adds another boundary to an agent workflow. Weave can help teams inspect the model decision, MCP tool invocation, latency, errors, and final task outcome together, which makes it easier to evaluate whether a connected capability is useful and safe.
Explore Token intelligence