Software for Agents
The intelligence layer your agent calls to make market decisions.
17 MCP tools across stocks, options, and crypto, with a public 30-day accuracy record your agent can cite. Drops into Claude Desktop, Cursor, Cline, or any MCP host.
Receipts the agent can cite
Install
Drop into any MCP host in 30 seconds
Stdio (Claude Desktop, Cursor, Claude Code):
{
"mcpServers": {
"marketintell": {
"command": "bunx",
"args": ["-y", "marketintell", "mcp"]
}
}
}HTTP / SSE (remote / web clients):
URL: https://api.marketintell.ai/mcp/sse Header: X-API-Key: mi_live_...
Hosts tested: Claude Desktop · Cursor · Cline · Continue · Claude Code.
The surface
17 tools, two transports, one rate-limit bucket
| Tool | Tier | Auth | Description |
|---|---|---|---|
| chat | free | optional | Auto-routed market chat (T1 fast / T2 CoT). Stocks, options, and crypto. |
| chat_deep | pro | required | T3 supervisor — parallel specialists + synthesis (~25–40s). Pro tier. |
| alpha_signals | free | optional | Latest published signals across all desks. Each links to a public post-mortem. |
| signal_performance | free | optional | Win-rate stats (1h / 4h / 24h windows). Same numbers as the public widget. |
| screener | free | optional | Trending + top gainers / losers across crypto + equities. |
| portfolio_list | free | required | List portfolios you own. |
| portfolio_create | free | required | Create a new portfolio with positions. |
| portfolio_analyze | free | required | PnL + risk + concentration analysis. |
| sessions_list | free | required | Your recent chat sessions. |
| billing_status | free | required | Plan, daily quota, monthly LLM cost, monthly cap. |
| health | free | none | Server uptime + provider circuit-breaker state. |
| register_challenge | free | none | PoW self-signup step 1 — issues a SHA-256 challenge. |
| register | free | none | PoW self-signup step 2 — provisions a Free-tier API key. |
| execute_propose_intent | pro | trade | Propose a trade to your connected broker — writes a pending intent you review + approve. No money moves here. |
| execute_approve_intent | pro | trade | Approve a pending intent and place the real order at the broker. Returns the executed-order audit row. |
| execute_list_intents | pro | trade | List your recent trade intents — pending, executed, rejected. |
| execute_read_portfolio | pro | trade | Read your connected broker account snapshot — equity, cash, buying power, open positions. |
The four execute_* tools place or read real broker orders, so they need a broker-connected key carrying the trade scope (paper by default; live mode requires an extra audit step) — not a plain API key.
Per-key MCP rate limit defaults to 30 calls/min in a separate bucket from REST. Each tool emits a span trace; if you set LANGFUSE_PUBLIC_KEY on your own deployment you can cite our spans directly.
Why agents need this
Trustworthy data + reasoning is the missing primitive
Generic LLMs can describemarkets but can't cite them, and the open-data APIs they reach for drift, return stale data, or 404 mid-call. Agents calling those endpoints look smart in a demo and break in production.
MarketIntell wraps live market data, derivatives, fundamentals, on-chain analytics, social signals, macro, and prediction markets — 27+ external sources behind a single MCP surface. Every response carries a citation trail. Every signal has a 24h + 7d outcome stamped on a public post-mortem URL the agent can return as a link.
Rate-limit bucket is per-key + separate from REST traffic, so a chatty autonomous agent won't starve a human Pro user on the same key. Cost-tracking is per-tool; you see exactly what each MCP call costs you in the billing_status tool.
Find us in
MCP catalogues we live in
Free tier, no credit card. Self-register a key by proof-of-work.
Get a key →