Kela MCP server

Plug Claude, ChatGPT, Cursor, or your own custom agent into your fund's live FundOS data. 47+ tools, MCP 2024-11-05 protocol, OAuth or API-key auth, full audit trail.

MCP 2024-11-05 SSE transport OAuth 2.0 + Bearer SOC 2 Type I

One-line setup for Claude Code

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

For other MCP clients

{
  "mcpServers": {
    "kela": {
      "url": "https://kela.com/mcp",
      "auth": { "type": "bearer", "token": "vdr_YOUR_API_KEY" }
    }
  }
}

What's available

Deal CRM

List, create, advance, and update deals. Run the screening agent.

LP CRM

List LPs, log meetings, draft follow-ups. Trigger the fundraising autopilot.

VDR & Diligence

Browse rooms, search documents, run the diligence agent over a room.

Transactions

Draft term sheets, manage closing tasks, extract CPs.

CFO & Risk

Compute waterfalls, run covenant checks, generate audit packs.

Compliance OS

List KYC records, filings, restricted securities, obligations.

Trading / OMS

Submit orders, check pre-trade compliance, list positions, FIX status.

Investors

List LPs, generate ILPA capital-call notices, distribution waterfalls.

Discovery files

/.well-known/mcp.json — machine-readable server manifest
/.well-known/agents.json — pre-built AI agents catalogue
/.well-known/oauth-protected-resource — RFC 9728
/.well-known/oauth-authorization-server — OAuth metadata
/llms.txt — overview for LLMs
/llms-full.txt — full integration reference
/CLAUDE.md — AI agent integration guide

Security model

External agents authenticate as a user (OAuth) or with a workspace API key.
Tool calls inherit the authenticating user's scope — no cross-org access.
Write-back actions queue for human approval; nothing auto-executes.
Every tool call is logged with trace_id + user_id + tool_name + duration in the CreditLedger.
Rate-limited per-user, per-feature (e.g. 10/min on Gemini-cost endpoints).

Sample agent flow

From a Claude conversation: "Find the LP with the largest commitment that hasn't been contacted in 60 days and draft a follow-up."

# 1. Claude calls fundos_list_lps → returns LP roster
# 2. Filters by commitment_amount desc + last_contact_at < now-60d
# 3. Calls fundos_get_lp(lp_id) for the top result
# 4. Calls fundos_run_agent("lp_fundraising_autopilot", lp_ids=[...])
# 5. Returns the drafted email to the user for approval
# 6. After approval, calls fundos_send_email — audit-logged.

Talk to us

If you're building an internal agent product for your fund and want a deeper integration walkthrough, book a 30-minute call.