AI ACCESS · MODEL CONTEXT PROTOCOL

Connect your AI agent to Marginal Thinking

Use the Marginal Thinking MCP to give an AI agent structured, read-only access to our public research on economics, politics, society, capital, productive capacity, resources, technology and global power.

https://www.marginalthinking.org/mcpOpen · read-only · no authentication

What this gives your agent

An AI model can answer broad questions on its own, but reproducing a serious research workflow still requires finding the right material, assembling context, comparing periods and connecting related developments. The Marginal Thinking MCP exposes that research layer directly to the agent.

Marginal Thinking maintains the research corpus. Your agent can then retrieve the relevant material and adapt it to your own question, company, portfolio, research problem or decision context.

Marginal Thinking → MCP → your agent → your decision

Connect

In an MCP-compatible client, add the remote server URL below. The server is public and stateless at the application layer, and all tools are read-only.

https://www.marginalthinking.org/mcp

Repository-level technical documentation remains available in MCP.md; this page is the stable public interface for users and agents.

What you can ask

Read the latest Marginal Thinking research on global capital flows and explain what matters for a Brazilian industrial company.

Compare the current critical-minerals research with previous Marginal Thinking reports and show what changed.

Build a 30-day timeline of Marginal Thinking research related to oil, sovereign yields and political risk.

Find research connected to Poland, industrial policy and European security, then summarize the second-order economic effects.

Available tools

search_research

Search the corpus by question or subject, with optional filters for program, topic, country, cadence, format, series and date range.

get_research

Retrieve canonical metadata and the public Markdown of a research item. A heading can be requested to return only one section.

latest_research

Return the most recent publications, optionally constrained by the same controlled metadata used by the site.

get_related_research

Find structurally related publications using programs, topics, geography, series and analytical dimensions.

get_timeline

Build a chronological view of research around a country, market, institution, technology, resource or other subject.

get_corpus_overview

Return the corpus-level coverage intelligence already derived automatically by the Marginal Thinking build.

Machine-readable resource

The server also exposes the public research catalog as an MCP resource:

marginalthinking://catalog

How it works

The MCP does not maintain a second knowledge base. It reads the same canonical research artifacts already produced by the Marginal Thinking publishing pipeline.

GitHub research repository
        ↓
existing Cloudflare build
        ↓
Markdown + canonical metadata
        ↓
MCP catalog
        ↓
/mcp on the existing Cloudflare Worker
        ↓
your MCP-compatible agent

Scope and limits

The MCP retrieves and organizes published Marginal Thinking research; it does not generate new Marginal Thinking analysis inside the server.
It does not browse the web, mutate the repository or call an LLM.
Time-sensitive claims should still be read together with the dates, evidence and cited sources in the underlying research.
When citing Marginal Thinking, prefer the canonical HTML URL. When extracting report text, prefer the Markdown returned by the MCP.

Why this matters

The internet made information abundant. AI makes it easier to work with that information. The remaining constraint is often the research layer in between: what to collect, how to connect it and how to preserve context over time. The Marginal Thinking MCP makes that layer directly usable by your own agent.