Back to Discover

tally-mcp

connector

worldclasstom

The risk officer for AI-assisted trading: enforced exit rules, cooling-off, and public receipts.

View on GitHub
0 starsSynced Aug 10, 2026

Install to Claude Code

/plugin marketplace add worldclasstom/tally-mcp

README

Tally — the risk officer for AI-assisted trading

Every trading connector makes your AI more capable. Tally makes it accountable.

Tally (tally.markets) is a trading-discipline layer that mounts inside Claude, ChatGPT, Claude Code, and any MCP-capable agent. Your AI is the analyst — it researches, monitors, and gathers evidence. Tally holds the record neither you nor the model can rewrite: pre-registered theses, exit rules that freeze before capital moves, a 24-hour cooling-off, an override ledger for rule-breaking exits, and a public receipt for every closed trade.

Tally never executes trades. Every broker and venue connection is read-only at the API level — no order placement, no custody, by architecture and by policy.

Connect

Server URL (Streamable HTTP · OAuth 2.0 with dynamic client registration):

https://tally.markets/api/mcp
ClientHow
ClaudeSettings → Connectors → Add custom connector → name it Tally, paste the URL. Leave the OAuth fields empty — Tally registers itself.
ChatGPTPlugins → New plugin → name it Tally, paste the URL as the Server URL, Authentication: OAuth. (Requires developer mode.)
Claude Codeclaude mcp add --transport http tally https://tally.markets/api/mcp, then /mcp to sign in.
Any MCP agentPoint it at the URL above; OAuth discovery does the rest.

Then say: "Get started with Tally, then run my checkup." The server teaches your AI the whole protocol on first contact.

Full walkthrough (skill, scheduled morning checkup, broker linking): tally.markets/connect

Tools

ToolWhat it does
get_startedReturns the discipline protocol — the AI teaches itself
get_checkupMorning briefing: active theses, deadlines, evidence, closed history and lifetime R
list_theses / get_thesisRead the journal
create_thesisPre-register a trade: statement, mechanism, risk budget, exit criteria (starts the 24h cooling-off)
arm_thesisFreeze the criteria after cooling-off — from here they fire, never edit
record_readingLog evidence against a criterion (the generic sensor primitive)
fire_criterionMark a rule objectively triggered, with cited evidence — one-way
close_thesisClose a trade; discretionary closes require a written justification, logged forever
get_receipt_linkThe public receipt for a closed thesis

Reads are safe to always-allow; writes are guarded by the protocol itself (cooling-off, criteria minimums, override justifications).

Proof

A real receipt — pre-registered thesis, venue-verified entry and exit, −0.22R, graded Process B / Outcome D, closing note verbatim: "Wrong on thesis, right on process."

How it compares

Market-data connectors inform your AI. Execution connectors let it trade. Journal connectors show it your past. Tally is the one built to constrain — it runs alongside all of them: tally.markets/compare

Privacy & security

  • OAuth per user; your AI sees your journal only, under a grant you can revoke any time.
  • Brokerage data (via SnapTrade) is read-only; broker credentials never touch Tally.
  • Details: privacy · terms · support@tally.markets

This repository contains documentation and the public server manifest only — the hosted server lives at tally.markets. No secrets, keys, or server source belong here, ever.

Rendered live from worldclasstom/tally-mcp's GitHub README — not stored, always reflects the source repo.

1 Install Method

NameDescriptionCategorySource
streamable-http remoteHosted streamable-http endpointmcp-serverhttps://tally.markets/api/mcp

0 Comments

Login required
Log in to post a comment or update on this repo.

No comments yet — be the first to share an update.