MCP server

@theneuralledger/mcp is a read-only Model Context Protocol server. It exposes intelligence as tools that an AI client can call — no client code required.

npx -y @theneuralledger/mcp

The binary is tnl-mcp. For client-specific configuration (Claude Desktop, Cursor, and others) see MCP connecting.

Running

stdio — the usual mode, for clients that launch a subprocess:

TNL_API_KEY=your_key npx -y @theneuralledger/mcp

HTTP — for a shared server several clients connect to:

import { createHttpServer, listenHttp } from '@theneuralledger/mcp';

const server = createHttpServer({ apiKey: process.env.TNL_API_KEY! });
await listenHttp(server, { host: '127.0.0.1', port: 8931 });

Bind to loopback and put a reverse proxy in front if it needs to be reachable — the server trusts anything that can connect to it.

Docker, when you would rather not install Node:

docker run --rm -i -e TNL_API_KEY=$TNL_API_KEY \
  ghcr.io/bekirdag/tnl-intelligence tnl-mcp

Embedding in your own server

import { createTnlMcpServer } from '@theneuralledger/mcp';
import { runStdio } from '@theneuralledger/mcp';

const server = createTnlMcpServer({
  apiKey: process.env.TNL_API_KEY!,
  userAgent: 'acme-assistant/2.1',
});

await runStdio(server);

registerResearchMcp(server, options) adds the research tools to a server you already own, if you are composing several MCP sources into one process.

Tools

Intelligence

Tool	Answers
tnl_latest_news	What has just been published
tnl_entity_intelligence	What is happening around an organisation or person
tnl_asset_intelligence	What is moving an asset
tnl_impact_path	How an event propagates to markets
tnl_explain_event	Plain-language explanation of a story
tnl_daily_risk_review	Today's risk picture
tnl_event_due_diligence	Structured checks on a single event

Research — slower, evidence-first, and they consume research quota:

Tool	Answers
tnl_deep_research	Multi-source investigation of a question
tnl_research_asset_exposure	Exposure of an asset to a theme or event
tnl_research_validate_event	Whether a claimed event holds up
tnl_research_compare_sources	Where sources disagree
tnl_research_operational_risk	Operational risk around an entity

Full argument schemas are in MCP tools.

Asking well

The tools are read-only, so the useful skill is picking the right one. In practice:

"Use tnl_asset_intelligence for KRW, then tnl_impact_path on anything it returns about policy rates."

beats asking a general question and hoping the client chooses. Naming the tool also makes the client's answer reproducible.

Research tools take longer and cost more quota than the intelligence tools. Reach for tnl_latest_news or tnl_entity_intelligence first, and escalate to tnl_deep_research only when the shallow answer is not enough.

Quota and safety

Every tool call is an API call against your plan. tnl status shows remaining quota; see plans.

The server is read-only — there is no tool that writes, deletes, or spends money, so an assistant cannot damage your account through it.