Show HN: I made an MCP Agents can reason every finance data the first time

Prereason first. Then answer.

Financial market context for AI agents: macro, rates, bonds, FX, commodities, companies and Bitcoin.

Docs · Briefings · Company briefings · Pricing · Get a key

Ask your AI where Treasury yields are. On its own, it reaches for a search result or a raw series and works out the rest by itself. When it prereasons, one get_context call brings back the context already worked out: the 30 year yield at 5.67% on 2026-10-07, the highest reading on the US Treasury's daily par yield curve since July 2001.

• The analysis, and the data behind it. Each briefing arrives with the reading done: the trend over several windows, where a value sits in its own history, a company's rank among its peers and, where read, its own guidance.
• Every figure dated, every source named. Your agent can say when a number is from and where it came from.
• One call, a whole briefing. A briefing gathers the series a question needs, so the turns go to the answer rather than to lookups.
• The markets in one place. Rates and Treasuries, the Fed, FX, commodities, macro, listed companies in the USA, Japan, Korea and India, and Bitcoin.
• It describes; it never advises. Briefings say what the data shows. They never say what to buy, sell or hold.
• Quick to set up. One block of config for Claude, Cursor, Codex or VS Code, through the npm bridge or the remote server. The catalogue tools need no key; for the rest, the bridge gives you one link to approve in your browser and the key arrives in the running session, and the remote server asks you to sign in.

PreReason gives an AI agent market context it can reason with. One call returns a briefing with the analysis already in it: a one line reading, trend direction over several windows, confidence scores, percentile ranks and correlations, and in the deeper briefings a regime label and a plain language narrative. The briefings cover macro, bonds, FX, commodities, Bitcoin, Ethereum and cross asset correlations. Company briefings cover listed companies in the USA, Japan, Korea, India, the UK, France, Austria, Belgium, Denmark, Finland, Iceland, Italy, Lithuania, the Netherlands, Slovenia and Slovakia, one company per call. The catalogue holds 33 live briefings and 293 individual metrics, among them Treasury yields, the latest FOMC statement, the macro release calendar, Treasury auctions, who holds US Treasuries, the Fed balance sheet, M2, net liquidity, the dollar, US crude and gas inventories, gold, silver, copper and wheat, oil, mining and agribusiness companies beside their commodity, AI compute and power demand, Bitcoin price and momentum, Bitcoin network and miner health, spot Bitcoin ETF flows and corporate Bitcoin treasuries. It is served over MCP (a remote server and an npm bridge) and over REST, as Markdown or JSON. Over MCP, the catalogue tools need no account, and a briefing needs a PreReason account, signed in through the connector or with an API key; a free account reads the free briefings. Over REST, a briefing takes an API key or, on the routes that take one, an x402 payment per call with no account.

Requires Node.js 18+, and nothing else: the bridge has no dependencies.

Add this to claude_desktop_config.json and restart Claude Desktop:

{
  "mcpServers": {
    "prereason": {
      "command": "npx",
      "args": ["-y", "@prereason/mcp"],
      "env": { "PREREASON_CLIENT": "claude-desktop" }
    }
  }
}

The catalogue tools work at once. To connect your account, tell Claude "connect me to PreReason", or ask for any briefing: either way Claude shows one link, https://www.prereason.com/claim/PR-XXXX-XXXX. Open it, sign in or create a free account, click Approve. The key arrives in the bridge on its own, is saved to ~/.prereason/credentials.json, and is never shown; ask Claude whether PreReason is connected and it names your plan. Nothing is created in your account until you click Approve, and the link is asked for when you need it, so it never runs out while Claude sits open. If your account already holds as many API keys as it allows, the answer says so instead of looping: revoke a key under Settings on prereason.com, then ask to connect again, or set PREREASON_API_KEY to a key you already have.

Config file location:

• macOS:~/Library/Application Support/Claude/claude_desktop_config.json
• Windows:%APPDATA%\Claude\claude_desktop_config.json
• Linux:~/.config/Claude/claude_desktop_config.json

Already have a key? Add it to the env block as "PREREASON_API_KEY": "pr_live_..." and the bridge never asks.

Option 2: Direct HTTP with an API key (Claude Code, Cursor, Windsurf, Codex, Gemini CLI, VS Code, scripts)

Clients that hold their own config can call the endpoint directly, with the key as a header:

claude mcp add --transport http prereason https://api.prereason.com/api/mcp --header "Authorization: Bearer YOUR_API_KEY"
{
  "mcpServers": {
    "prereason": {
      "type": "http",
      "url": "https://api.prereason.com/api/mcp",
      "headers": { "Authorization": "Bearer YOUR_API_KEY" }
    }
  }
}

Windsurf uses serverUrl instead of url; Gemini CLI uses httpUrl; Codex uses url plus bearer_token_env_var in config.toml. No key yet? Point the client at the endpoint without a header to browse the catalogue, then create a key on the website (below) and add the header.

Add PreReason as a custom connector, choose "No sign-in", and where the Request headers section is available add Authorization with the value Bearer YOUR_API_KEY (the word Bearer and the space are part of the value). Sign in support for connectors is being re-tested and is not offered until it is verified.

Three ways, all free. The hosted server never hands out a key inside a session and never asks for one.

Through this bridge. Run it with no key set: when you ask your assistant to connect PreReason, or an answer needs a key, it asks PreReason for access on your behalf and shows one approve_url, in its log and in the answer. Open it, sign in or create your account, click Approve, and the bridge saves the key, attached to your account and named Agent: <client_name>.

From code, for an agent with no browser. POST https://api.prereason.com/api/agent/claims (no auth), show the human the approve_url, then poll GET https://api.prereason.com/api/agent/claims/{claim_code} with Authorization: Bearer <claim_token> until status is approved. Docs: prereason.com/docs#agent-access.

On the website. Sign up at prereason.com/signup, then Dashboard > Settings > API Keys. Keys start with pr_live_.

The npm bridge and the Claude Desktop extension add one tool of their own, connect_account, which the bridge answers itself and the remote server does not offer: with no key yet it returns the one link to approve, and once approved it says the account is connected and on which plan. It never shows a key and never asks for one.

equities.company reads one listed company per call, composed from its own reports (filings with its home market's regulator; for India, its own published results), each figure dated and sourced: USA companies on Basic and above, Japanese, Korean and Indian companies on Pro and above. It describes; it never says what to do.

• Name the company withentity : a USA ticker or CIK (NVDA ), a Japanese securities code (7203 ), a Korean stock code (005930 ), an Indian company's ISIN (INE018A01030 ) or the LEI of a company filing in the UK or of a European company (2138002P5RNKC5W2JZ46 ).
• Not sure of the code? Calllist_briefings withsearch , for example"Toyota earnings" : the answer lists the matching covered companies, each with theentity to pass.
• Read one part withblock , one of the 11 below, or leave it out for the whole briefing. A part counts one call; the whole briefing counts one for each part it carries. Parts are served for today only.

get_context(briefing="equities.company", entity="7203")
get_context(briefing="equities.company", entity="NVDA", block="valuation")

Once connected, try prompts like:

• "Connect me to PreReason"

• "Show me the macro snapshot"

• "Where are Treasury yields, and what did the Fed last decide?"

• "What did NVIDIA report, and how does it rank among its peers?"

• "How is Bitcoin moving with stocks and the dollar?"

• "List available briefings"

• Ensure Node.js 18+ is installed: node --version

• Check your API key starts with pr_live_

• Fully quit Claude Desktop (system tray > Quit) and reopen

• Restart Claude Desktop after editing config

• Verify JSON syntax: node -e "JSON.parse(require('fs').readFileSync('path/to/config','utf8'))"

If you still see this error, ensure you're using the env block (not --header args) as shown in Quick Start above. If the issue persists, install globally and use node:

1. Run: npm install -g @prereason/mcp
2. Use this config (replace YOUR_USER with your Windows username):

{
  "mcpServers": {
    "prereason": {
      "command": "node",
      "args": [
        "C:\\Users\\YOUR_USER\\AppData\\Roaming\\npm\\node_modules\\@prereason\\mcp\\bin\\cli.js"
      ],
      "env": {
        "PREREASON_API_KEY": "YOUR_API_KEY"
      }
    }
  }
}

• list_briefings ,list_metrics , andget_health work without a key
• get_context andget_metric require a valid API key, andget_changes a key on the Basic or Pro plan
• Get a free key at prereason.com/signup

If your client supports remote HTTP servers, use Quick Start Option 2 above. The stdio bridge package is only needed for clients that require stdio transport (e.g. Claude Desktop).

npx @prereason/mcp

npx @prereason/mcp --login

npx @prereason/mcp --logout

PREREASON_API_KEY=pr_live_... npx @prereason/mcp

PREREASON_CLIENT=claude-desktop npx @prereason/mcp

npx @prereason/mcp --header "Authorization:Bearer YOUR_API_KEY"
npx @prereason/mcp --credentials-file /path/to/credentials.json
PREREASON_URL=https://custom.endpoint/mcp npx @prereason/mcp

npx @prereason/mcp --help

Key precedence: PREREASON_API_KEY, then --header, then the credentials file (~/.prereason/credentials.json, or PREREASON_CREDENTIALS_FILE, or --credentials-file), then the claim flow. The file holds the key and which claim issued it, never a claim token. On macOS and Linux the directory is created 0700 and the file 0600; Windows has no mode bits, so the file relies on your profile directory's permissions like every other credential store there.

mcpb/manifest.json describes the same bridge as a single click Claude Desktop extension, key optional. To build the bundle: npm install --omit=dev, then npx @anthropic-ai/mcpb pack . from the package directory, and install the resulting .mcpb by double clicking it. Submission to the Claude directory goes through the desktop extension form and is a publisher decision.

The bridge ships its own transports and installs nothing. npm ls on it is one line, npx @prereason/mcp fetches one 23 KB tarball and starts, and the code a security review has to read is the code in this repository.

It used to depend on @modelcontextprotocol/sdk for two classes, a stdio transport and a Streamable HTTP client. That pulled in 91 packages and 25 MB on disk, nearly all of it the SDK server half: Express, Hono, CORS, a rate limiter, an OAuth client and a schema validator, none of which a relay ever calls. lib/stdio.js and lib/streamable-http.js replace the two classes the bridge used, keep their framing and their callbacks, and are covered by the suite under test/.

See prereason.com/privacy for data handling practices.

MIT