Mdtero

Use Mdtero to turn one paper into a structured Markdown research package for downstream OpenClaw work.

Start in Mdtero Account for keyword discovery and API-key management.

ClawHub install only sets up the agent skill. It does not install the local helper.

When To Use

Use this skill when the user wants to:

• - parse a paper from a DOI, supported publisher page, or open URL into a Markdown research package
• translate an already parsed paper without losing the original research structure
• check task status and download paper_md, paper_bundle, optional paper_pdf, or INLINECODE3

Required Setup

1. 1. Check whether MDTERO_API_KEY is already available.
2. If it is missing, tell the user to generate one at https://mdtero.com/account.
3. Explain that keyword discovery and API-key management stay in Mdtero Account.
4. Ask them to set it in their local shell, for example:

CODEBLOCK0

1. 5. Authenticate requests with:

CODEBLOCK1

1. 6. If the user wants Elsevier or ScienceDirect full-text retrieval, also check ELSEVIER_API_KEY and explain that it is separate from MDTERO_API_KEY.
2. If an Elsevier DOI is missing the local helper or ELSEVIER_API_KEY, tell the user exactly what is missing instead of guessing.
3. Explain that ELSEVIER_API_KEY helps the user's local acquisition step. It does not enable direct server-side POST /tasks/parse for 10.1016/... inputs.

Before Use

• - Mdtero API usage requires MDTERO_API_KEY. Elsevier or ScienceDirect full-text retrieval may also require ELSEVIER_API_KEY.
• Tell the user that ClawHub install only sets up the agent skill, while the local helper remains a separate install on their own machine.
• Parse and translate requests send paper content or paper-source content to https://api.mdtero.com, so do not use this workflow for sensitive or proprietary manuscripts unless the user accepts that boundary.
• If the user needs the local helper, tell them to download the installer, review it locally, then run it on their own machine. Do not recommend piping a remote script directly into the shell.
• If the user wants a fuller install or helper walkthrough, point them to https://mdtero.com/guide or https://api.mdtero.com/skills/install.md.

Workflow Rules You Must Preserve

• - PDF is an optional artifact. Prefer the Markdown package first and only fall back to paper_pdf when the workflow truly requires it.
• If the user starts from a local PDF, explain that Mdtero currently defaults to GROBID; Docling and MinerU remain fallback options.
• For Elsevier and ScienceDirect, local acquisition should stay on the user's own machine through the local helper or browser extension.
• Direct POST /tasks/parse is for open inputs. Elsevier and ScienceDirect full text must go through local acquisition first, then POST /tasks/parse-fulltext-v2 or POST /tasks/parse-helper-bundle-v2.
• If an Elsevier parse only returns the abstract, ask whether the user is on a campus or institutional IP.
• For Elsevier papers, prefer the raw DOI form such as 10.1016/j.energy.2026.140192.

Local Helper

When the user needs local acquisition for Elsevier or ScienceDirect, tell them to open the official Mdtero guide at https://mdtero.com/guide or the install handoff at https://api.mdtero.com/skills/install.md.

Tell them to download the helper installer, review it locally, then run it on their own machine. Explain that the installer auto-detects python3, python, or node, and exposes the mdtero-local command without requiring extra packages. Do not recommend piping a remote script directly into the shell.

If they need the exact helper handoff, point them to https://api.mdtero.com/helpers/install_mdtero_helper.sh and tell them to save it locally before running it.

Explain that mdtero-local parse and mdtero-local translate short-wait by default. Fast tasks may already come back completed; slower tasks still return a pending task_id. The user can force pure async mode with mdtero-local parse --no-wait ... or mdtero-local translate --no-wait ....

Explain that the public API contract itself is still task-based: POST /tasks/parse and POST /tasks/translate return task_id, while GET /tasks/{task_id} remains the stable completion path.

Parse A Paper

Open inputs can go directly to POST /tasks/parse.

CODEBLOCK2

The response returns a task_id.

For Elsevier or ScienceDirect full text, do not send the DOI straight to POST /tasks/parse. Use the local helper or browser extension first, then upload the local helper bundle:

CODEBLOCK3

If the API says Elsevier or ScienceDirect inputs must be acquired locally first, that is the expected behavior rather than a setup failure.

Translate A Parsed Markdown File

Use POST /tasks/translate with the server-side Markdown artifact path returned by a succeeded parse task in result.artifacts.paper_md.path. Do not substitute an arbitrary local file path.

CODEBLOCK4

The response returns a task_id. Poll it, then download translated_md.

Check Task Status

Poll GET /tasks/{task_id} until the task becomes succeeded.

CODEBLOCK5

The result object returns artifact metadata rather than the file body.

Download Artifacts

Download file contents from the download route.

CODEBLOCK6