MCP server

Market research inside your MCP client.

Ask market questions from Claude Code, Cursor, VS Code, or any other MCP client. There's one tool, ask_neuroapi, and it returns a researched answer on stocks, crypto, forex, commodities, or macro.

https://api.neurobro.ai/api/v1/mcp
[01] INSTALL

Connect your client.

You'll need an API key from a paid plan. Each config reads it from an environment variable or a secret prompt, so it stays out of your repo.

Run this in your terminal.

$ claude mcp add --transport http neuroapi https://api.neurobro.ai/api/v1/mcp --header "X-API-Key: $NEUROAPI_KEY"

Install in one click, or add this to ~/.cursor/mcp.json.

1{
2 "mcpServers": {
3 "neuroapi": {
4 "url": "https://api.neurobro.ai/api/v1/mcp",
5 "headers": { "X-API-Key": "${env:NEUROAPI_KEY}" }
6 }
7 }
8}

Install in one click, or add this to .vscode/mcp.json. VS Code asks for the key once and keeps it in secret storage.

1{
2 "servers": {
3 "neuroapi": {
4 "type": "http",
5 "url": "https://api.neurobro.ai/api/v1/mcp",
6 "headers": { "X-API-Key": "${input:neuroapi-key}" }
7 }
8 },
9 "inputs": [
10 {
11 "id": "neuroapi-key",
12 "type": "promptString",
13 "description": "NeuroAPI API key",
14 "password": true
15 }
16 ]
17}

The custom connector screen has no field for an API key, so connect through the mcp-remote bridge. Keep X-API-Key:${NEUROAPI_KEY} as one token, with no space after the colon.

1{
2 "mcpServers": {
3 "neuroapi": {
4 "command": "npx",
5 "args": [
6 "-y",
7 "mcp-remote",
8 "https://api.neurobro.ai/api/v1/mcp",
9 "--header",
10 "X-API-Key:${NEUROAPI_KEY}"
11 ],
12 "env": { "NEUROAPI_KEY": "neuro_..." }
13 }
14 }
15}

Any client that speaks Streamable HTTP and can send a header works with these settings. For stdio-only clients, use the mcp-remote config from the Claude Desktop tab.

1URL https://api.neurobro.ai/api/v1/mcp
2Header X-API-Key: neuro_...

To check it worked, list the tools in your client. You should see ask_neuroapi and nothing else.

[02] THE TOOL

One tool, five inputs.

InputWhat it does
promptThe market question. This is the only required input.
modefast, smart, or max. Defaults to smart.
message_historyEarlier turns as {role, content} pairs. Your client keeps the conversation.
system_promptExtra instructions for the agent. It can't override the agent's guardrails.
output_schemaA JSON Schema. The agent returns a validated object instead of prose.

The response has answer (or output if you passed a schema), mode, request_id, and usage. Tool calls return one complete answer. To stream tokens as they're written, call POST /agent/ask with stream=true.

Pick a depth per call. Each one costs a set number of quota units.

ModeBest forUnitsPlans
fastQuick answers from core market knowledge1All
smartStronger reasoning with deeper context2All
maxDeep multi-asset research8Pro, Enterprise
[03] AUTH AND LIMITS

Same key and quota as the REST API.

Authentication
Send your key as X-API-Key. Clients with only a bearer field can use Authorization: Bearer on this endpoint. A key in the URL returns 400.
Billing
Each call uses your plan's monthly quota at the units above. Failed calls are free.
Rate limits
Your plan's per-minute limit applies to each call, on top of per-IP limits on the endpoint.
Runs at once
Starter 2, Pro 5, Enterprise 10, across all your keys and MCP.
Access
The server exposes ask_neuroapi and nothing else. It can't reach your dashboard or billing. Browser requests from origins not on the allow-list get 403.
[04] FAQ

Before you connect.

01Can I try it for free?

Yes. The agent behind the API is free in Neurobro Web and in Neurobro Mobile for iOS and Android. The API and MCP server need a paid plan, so when you want it inside your own product, grab an API key.

02Which clients work?

Claude Code, Cursor, and VS Code connect directly. Claude Desktop and other stdio-only clients connect through the mcp-remote bridge. Any client that speaks Streamable HTTP and can send an X-API-Key header will work.

03How is this different from a market data MCP server?

A market data server hands your model prices and datasets, and the model still has to make sense of them. ask_neuroapi does the research itself and returns a finished answer from one tool.

04Can I stream the answer?

Not over MCP, where tool calls return one complete answer. To stream, call POST /agent/ask with stream=true.

[05] START

Add NeuroAPI to your client.