Model Context Protocol
An open protocol for connecting language models to tools and data sources, and the interface through which CivBench exposes Civilization VI to an agent.
Standard · 2024Sources & confidence: High confidence
Checked 2 October 2026 · AI-assisted review
Versioned protocol specification and supported interaction primitives.
- Model Context Protocol specification, 2025-11-25
Resources, prompts, tools and client/server capabilities; does not prescribe a one-observation-at-a-time agent architecture.
- Wikidata entity API: Model Context Protocol
Inspected entity label and description for identifier matching only; does not validate the work’s argument.
What this check doesn’t establish
- The protocol was checked, not the civ6-mcp implementation. The 2024 date denotes original introduction; this review inspected the named 2025 specification.
Confidence applies to the scope above, not every claim made by the underlying source. How confidence works
The protocol civ6-mcp implements supports tools, resources and prompts. It does not itself require an agent to perceive only one explicitly requested tool response at a time.
The essay’s “sensorium effect” concerns the particular interface and agent loop used in its experiment. Generalising from that setup to every MCP system would confuse an implementation choice with a protocol requirement.
In the essays
- I Gave an AI a Civilization to Run. It Built a Nuke.
Read the essay
Projects
- civ6-mcp
- Parliament MCP
Implemented by
Record history
Catalogue source: knowledge/works/model-context-protocol.md
Generated by claude-code/claude-opus-5 on 2026-07-25