FundOS Developer Docs

Connect any AI agent to live fund data in one command. FundOS exposes deals, LPs, covenants, documents, and CFO functions through a unified MCP server — with OAuth handled automatically.

What FundOS is

FundOS is an AI-native operating system for fund managers — hedge funds, private equity, private credit, venture capital, and family offices. It exposes 47 MCP tools across 11 modules, each with a consistent read/write permission model and credit-based billing.

Every write tool produces a proposed action that waits for human approval in the FundOS UI. Nothing auto-executes without a GP clicking Approve.

Connect in one command

claude mcp add fundos https://kela.com/mcp

For SSE-only clients: https://kela.com/mcp/sse
For Bearer-token API access: Authorization: Bearer vdr_<your-key> against https://kela.com/api/v1/*

Key discovery endpoints

URLDescription
/CLAUDE.mdFull agent guide and engineering reference
/llms.txtShort FundOS overview for LLMs
/llms-full.txtFull developer + integration reference
/.well-known/mcp.jsonMCP server discovery
/.well-known/oauth-protected-resourceOAuth resource metadata
/.well-known/oauth-authorization-serverOAuth authorization-server metadata

Permission model

Every tool is either a read (safe to call freely) or a write / AI-heavy (requires a human to approve the resulting action). The table below summarises what each tier does and costs.

TierWhat it doesCost
read_fast Single-table reads — list deals, list LPs, etc.1 cr
read_multi Cross-table reads — agent context briefing2 cr
read_compute Compute-heavy reads — pricer, waterfall, covenant check5 cr
write Writes that create or mutate records (human approval required)10 cr
ai_light Gemini AI calls — VDR analysis, ODD generation25 cr
ai_heavy Long-form AI — CIM generation50 cr

Human approval loop

Write tools never execute automatically. When an agent calls fundos_create_deal, for example, the MCP server creates a proposed action and returns metadata about it. The GP reviews and approves the action at /fundos/agent/runs/<id>. Only then does the side effect execute. This means agents can draft, plan, and propose — but they cannot accidentally move money, change records, or share data without a human in the loop.

Start here: Call fundos_get_agent_context at the beginning of any multi-step workflow. It returns a structured session briefing — pending actions, recent tool calls, active deals, LP changes, open risk alerts, and upcoming capital calls — for 2 credits.

Quicklinks