Artha Council MCP server

Artha Council

<!-- mcp-name: io.github.akira231097/artha-council -->

Artha is an AI-assisted equity research and portfolio automation system. It narrows a broad US-stock universe, ranks opportunities, asks independent analyst roles to debate the strongest candidates, monitors active investment theses, and uses deterministic broker gates before an order can proceed.

Version 1.4 hardens the MCP server with source/build provenance for portable local or OAuth-protected remote access. It supports the existing US workflow and market-aware broker adapters for Indian cash equities without treating US research as India data. India adapters provide broker-verified instrument lookup and fail-closed, whole-share delivery limit execution with current static-IP requirements.

Created and maintained by Sarath (@akira231097).

See the design notes and architecture diagram for a visual walkthrough of the research, Council, execution, and audit boundaries.

Artha is research software, not financial advice. Public defaults cannot place

live trades. Anyone enabling broker integration is responsible for reviewing

the source, their configuration, broker terms, and applicable requirements.

Why Artha Exists

Most stock screeners stop after ranking tickers. Artha treats investing as a stateful operating system:

1. Find candidates from a large universe.
2. Verify data quality and execution feasibility.
3. Give scarce Council attention to the strongest candidates.
4. Separate investment judgment from order execution.
5. Monitor every held thesis and send material changes to a sell Council.
6. Reconcile intended actions against broker state.
7. Preserve evidence and state transitions for audit.

System Pipeline

Market data providers
  -> universe and promotion funnel
  -> investment sanity checks
  -> broker-aware candidate router
  -> opportunity scout
  -> fundamental / technical / risk analysts
  -> CIO synthesis
  -> execution officer
  -> deterministic broker review and placement gates
  -> reconciliation, thesis tracking, Telegram, supervisor

The sell side runs independently of new-buy capacity:

Portfolio snapshot
  -> position and thesis monitor
  -> deterministic stop / invalidation triggers
  -> sell Council for judgment decisions
  -> execution officer
  -> broker review and exact-order placement gate
  -> post-fill reconciliation

Important Design Rules

Research is not execution

The Council answers: "Is this investment attractive?"

The execution layer answers: "Can this exact order be placed safely now?"

A good company can be temporarily unbuyable because the quote is stale, the spread is too wide, the price moved above the approved cap, or broker review is incomplete. Those conditions do not rewrite the investment thesis.

Missing broker proof blocks an order

Money-moving paths fail closed. A model statement that a check passed is not enough; Artha requires the decisive structured broker output.

Portfolio monitoring continues when buys pause

Position limits and invested-capacity limits pause new buys only. Monitoring, sell review, reconciliation, health checks, and alerts continue running.

Major Components

Shortened. The full README is on GitHub.