EssenceScholar MCP Server

MCP server exposing the EssenceScholar deep research pipeline and paper workflows as agent tools.

Tools

Research pipeline

• deep_research — full literature review pipeline (2–8 min, searches EconPapers/arXiv/SSRN, enriches top papers, generates a structured draft)
• deep_research_chat — read-only follow-up on an existing session (ask questions, dig deeper)
• deep_research_expand — fetch more papers into an existing session via new searches (2–5 min)
• list_deep_researches — list all past research sessions
• get_deep_research — retrieve any past draft by ID

Search & library

• search_econpapers — raw EconPapers search, returns JSON paper objects
• search_ssrn — raw SSRN search, returns JSON paper objects
• download_paper — download a found paper (EconPapers/SSRN/arXiv/DOI URL) into your library and get its paper_id. SSRN refuses servers outright, so those are parked, opened in the reader's own browser, and — with the Essence Scholar extension 1.9.10+ — downloaded and imported by the extension while the call waits (up to ~90s); {"status": "parked"} only when that did not finish in time. Not an error, not billed; ESSENCESCHOLAR_MCP_NO_BROWSER=1 skips the browser step
• upload_paper — upload a local PDF into your library and get its paper_id
• register_paper_text — register a local PDF by extracting its text client-side (no server OCR); paper_id derived from the sha256 of the PDF bytes, so re-registering is a no-op
• list_user_papers — list papers in your library (resolve paper_ids; optional title filter)
• search_paper_content — server-grade RAG retrieval over one of your papers (the same hybrid BM25+semantic ranking a workflow step uses); returns ranked chunks + joined text
• get_paper_context — the server-held {{variable}} dict a workflow run would substitute (research_interests, missing/top/trending lit, author profiles, …) plus the paper's section inventory

The whole product, one door (new in 1.0.0)

• capabilities — what Essence Scholar can do, with prices, read from the app's own catalog. Free
• route — classify a sentence the way the app does and see what it WOULD do, with nothing run. Free
• quote — price one capability over an exact payload and mint a ten-minute confirm token. Free
• act — run what was quoted, once the user has agreed. The same consent gate the app uses

Long jobs — start, then poll (new in 1.0.0) A deep submission check takes 30–40 minutes and is billed when it starts. These make a paid run recoverable instead of trapping the report inside the call that launched it.

• get_submission_check — the report if it has finished, the live status if not
• get_deep_review — a Verified Deep Review by run id
• get_run / list_runs — durable handles for anything the app or act started

Notebooks, conferences, authors (new in 1.0.0)

• list_notebooks, create_notebook, add_papers_to_notebook, notebook_overview — a notebook is the product's answer to "several papers at once": shared retrieval, one overview, one chat
• list_conferences, get_conference, import_conference — conference programmes as agendas
• paper_authors, author_profile, author_deep_analysis — the evidence-linked researcher profiles
• park_paper — hand a paper the server may not fetch (SSRN) to the reader's own browser, where the extension imports it

Workflows

• list_paper_workflows — list your routines, every scope (paper, notebook, conference)
• run_paper_workflow — run a workflow (by ID) on a paper in your library (server-executed)
• run_workflow_on_pdf — upload a local PDF and run a workflow on it in one call
• get_workflow_definition — fetch a workflow's raw declarative DAG (steps/prompts/deps) as JSON
• compile_workflow_skill — convert a workflow into a portable skill (target=claude|codex|gemini) your own agent runs, instead of executing it server-side
• create_workflow — author a workflow FROM your agent: design the step DAG in conversation, push it to Agent Studio (returns validation warnings before any run)
• update_workflow — refine an existing workflow in place (steps/prompts/synthesis)

Typical flow: search_econpapers / search_ssrn → download_paper → run_paper_workflow. Or, to reuse a workflow as an agent skill: list_paper_workflows → compile_workflow_skill → save the returned SKILL.md. Compiled skills run client-side (Mode B): register_paper_text (local PDF) or list_user_papers → get_paper_context for the {{variables}} → search_paper_content per step for evidence.

Setup

Add to your Claude config (~/.claude/settings.json):

{
  "mcpServers": {
    "essencescholar": {
      "command": "uvx",
      "args": ["essencescholar-mcp"],
      "env": {
        "ESSENCESCHOLAR_API_KEY": "sk_live_..."
      }
    }
  }
}

Get your API key from essencescholar.com.

To install as a Claude Desktop extension instead, download the packaged .mcpb bundle from a release and open it with Claude Desktop — it will prompt you for your API key via manifest.json's user_config.

Privacy Policy

This connector sends the arguments of whichever tool you or Claude explicitly invoke (search queries, paper content/text, workflow definitions, research topics) to the EssenceScholar backend, associated with your EssenceScholar account. It has no access to Claude's conversation history, memory, or files beyond what a tool call explicitly names, and nothing runs unless a tool is called. See the full policy, including the Claude / MCP Connector section, at:

https://essencescholar.com/privacy-policy