Human-directed research · Agent-accelerated

WebMCP Research Workbench

The agent gathers. You decide what counts.

A shared research workspace where a WebMCP-enabled agent searches real scholarly sources, proposes evidence, and drafts a cited brief — while you control what counts as evidence and what gets approved.

Real OpenAlex research · human-controlled evidence · source-linked conclusions

Live workspace position

Research Cycle

YOUR TURN
  1. DefineHumanCurrent
  2. ResearchAgentUpcoming
  3. CurateHumanUpcoming
  4. SynthesizeAgentUpcoming
  5. ApproveHumanUpcoming

YOUR TURN

Define the research mission

Set the question, audience, and evidence limit below. The agent cannot begin until the human-owned mission exists.

Next: Save the mission. The Research Cycle will move to Agent's Turn — Research.

Who does what

Human judgment, agent acceleration

Human

Set the research question, decide which sources become evidence, edit the draft, review it, and give final approval.

Agent

Searches OpenAlex, inspects source records, proposes evidence for your review, and drafts a brief from the evidence you accepted.

WebMCP

Gives the agent structured access to the same workspace and its declared capabilities instead of requiring it to scrape the screen or imitate human clicks.

The website does not run an embedded AI model; a WebMCP-enabled agent uses the capabilities exposed by the workbench.

Human authority

Research Mission

Human-owned

The mission is the question your research works toward — it guides searches, evidence proposals, and the final brief.

Example: How effective are heat-pump retrofits at reducing residential energy use in cold climates?

Audience/context example: Briefing for a city sustainability team evaluating retrofit incentives.

Only the visible human interface can set or change the mission. Reset before changing a mission that already has evidence or a brief.

Agent-first discovery

Optional human source verification

OpenAlex only

The agent performs discovery during the Research stage through WebMCP.

It may search OpenAlex in Keyword and Semantic modes and inspect candidate records. Selected candidates enter the shared workspace under Agent Proposals for your review. You do not need to repeat the agent's searches.

Open manual search and source inspectionOptional — independently verify or explore OpenAlex sources.

KeywordBest when you know the names, terms, acronyms, or phrases you want to search for.

SemanticBest when you know the idea you're researching but relevant work may describe it using different words.

Results are limited to 5 compact records.

Source details

Choose one result to inspect its known metadata.

Agent handoff

Agent Proposals (0)

Human decision required

Nothing is waiting for review yet. When an agent proposes promising sources, they'll appear here for you to accept or reject. Only accepted sources become evidence.

Human-controlled membership

Accepted Evidence (0)

Human accepted

No human-accepted evidence yet.

Synthesize

Evidence Brief

No agent draft yet. The agent can draft only after at least one source is human-accepted.

Collaboration trail

Collaboration log

See the human decisions and agent actions that shaped the workspace.

No workspace activity yet.