MCP Endpoints

Initializes a Denario project on your machine: it creates the project directory and data_description.md. If git and gh are available it also makes the first commit and pushes a repo under your own identity — optional, and skipped otherwise. This tool is part of the client already installed on your machine — it never contacts the server, so your data description stays local.

Ask Claude:

Use denario_setup to create a project at ~/research/oscillators
with this data description: "..."

Generates a research idea for an existing project. The client sends your local data_description.md to the server, which runs the private idea pipeline and returns idea.md as data. The client writes it into your project and commits it. Takes ~2–3 minutes.

Ask Claude:

Now run denario_idea on ~/research/oscillators.

Underlying HTTP endpoint

The client calls this for you; shown for reference.

POST https://denario-mcp-engine-tnf7nmfboa-uc.a.run.app/v1/compute/idea
Authorization: Bearer denario_live_...
Content-Type: application/json

{
  "project_iteration": 0,
  "inputs": { "Iteration0/input_files/data_description.md": "..." },
  "options": {}
}

Generates the research methodology. The client sends your local data_description.md and idea.md to the server, which runs the private methods pipeline and returns methods.md as data. Run denario_idea first. Takes ~2–3 minutes.

Ask Claude:

Now run denario_methods on ~/research/oscillators.

Evaluates the quality of an iteration's results and decides what's next: Methods module (iterate — run denario_methods at the next iteration) or Paper module(done — write the paper). The client uploads the iteration's input markdown files; the server returns the evaluator's report/feedback plus the decision.

Ask Claude:

Run denario_evaluate on ~/research/oscillators and tell me whether to iterate.

Classifies the generated paper into arXiv categories. The client sends the paper section files (paper_output/temp/{Title,Abstract,Methods}.tex) to the server and receives classification.json back. Requires a paper to have been generated first.

Ask Claude:

Run denario_classify on ~/research/oscillators.

Writes the full scientific paper and compiles it to PDF. The client uploads the iteration's inputs (data_description, idea, methods, results, and any plots); the server runs the private writer pipeline, compiles the LaTeX, and returns paper.tex + paper.pdf as data, written into your project. This is a long step (~10 min); the client waits for it to finish. Run denario_methods first and make sure results.md exists.

Ask Claude:

Run denario_paper on ~/research/oscillators.

Publishes a finished project. It rewrites README.mdfrom the paper's title and abstract, makes a commit and pushes, and enables GitHub Pages — all under your own git/gh identity, so it never contacts the server. Run denario_paper first (it needs paper.tex in the project root); denario_classify is optional and adds arXiv categories to the page. Idempotent — safe to re-run.

Ask Claude:

Run denario_publish on ~/research/oscillators.

Reports project status by reading files on your machine: which iterations exist, whether each is complete (idea, methods, results, feedback), and the best iteration. Never contacts the server.

Reads a file from your local project directory (e.g. an idea.md or report.md) and returns its contents. Never contacts the server.

Lists files in a local project directory matching a glob pattern (recursive). Handy for seeing what a step produced. Never contacts the server.