Protologue

What Protologue MCP server does

Protologue MCP server provides an MCP access point for the Protologue reference, a sourced taxonomy covering prompting and LLM techniques. The project describes 215 entries across 11 branches. Each entry has a permanent PTL- identifier, a standalone definition, possible synonyms, links to broader, narrower, and related concepts, and associated primary sources.

The reference is intended for use by both people and software. Its site publishes schema.org DefinedTerm JSON-LD, llms.txt, llms-full.txt, entry-level Markdown and JSON, and complete JSON, CSV, SKOS JSON-LD, and Turtle exports. Entries also include citation formats and metadata for scholarly discovery.

How it works

The MCP deployment runs as a Cloudflare Worker named protologue-mcp. Its published endpoint is https://mcp.protologue.com/mcp. The repository is connected to the official MCP registry under com.protologue/protologue.

The source taxonomy is maintained in YAML files under data/, with categories in data/categories.yaml and term records in files named data/terms-*.yaml. A Python generator builds the static site and related machine-readable output. Changes are proposed through pull requests; merging a change to main triggers the site and MCP deployment process described in the repository.

Setup and configuration

The supplied material documents a hosted endpoint rather than a local MCP installation procedure. Clients that can connect to a remote MCP endpoint can be configured with https://mcp.protologue.com/mcp.

For contributors working on the underlying project, local site generation requires Python and the PyYAML package. The documented commands are:

pip install pyyaml
python3 build.py

The build writes generated content to site/. The MCP Worker is stored in the mcp/ directory, but the excerpt does not provide local Worker commands, configuration variables, or authentication requirements.

Tools and capabilities

The Protologue MCP server is suited to agents that need reference material about prompting and LLM techniques, especially when source attribution matters. The project supports:

• Definitions that can stand alone when quoted
• Permanent identifiers for taxonomy entries
• Synonym, broader, narrower, and related-term links
• Primary-source references, including papers and documentation
• Structured representations for machine processing
• Plain citations, BibTeX, and Google Scholar metadata
• Benchmark descriptions that omit changing scores and rankings

The supplied README does not enumerate the names, arguments, or return schemas of individual MCP tools. Confirm those details from the live server or its MCP registration before building tool-specific logic.

Limitations and notes

Protologue MCP server should be treated as a reference taxonomy, not as a live benchmark leaderboard. Benchmark entries describe what a benchmark measures, how it is scored, the prompting setup used for reported scores, and known limitations; they intentionally do not include scores or rankings.

The project states that the dataset contains 215 entries and 214 sources, with source checks recorded as of 2026-10-14. Those counts and review dates can change as the repository evolves. The material also identifies the project license as CC BY 4.0, which permits reuse with attribution.

No individual MCP tool list, client-specific setup instructions, environment variables, or local server startup commands are provided in the supplied excerpt. Client compatibility and exact request behavior should therefore be verified against the deployed endpoint or current registry metadata.