Skip to content

Haven Financing MCP

The decision to borrow against a portfolio is made once, in a hurry, usually against one number somebody quoted from memory. Margin rates differ by brokerage, a securities-backed line is priced off a different curve, and the broker's own simulator is behind a login. Haven publishes the terms as six read-only tools on a keyless endpoint, so the comparison can be run in the conversation instead of in a spreadsheet someone rebuilds.

Server type:  remote Streamable HTTP, keyless
Endpoint:     https://app.haven.finance/api/mcp
Server:       haven-financing 1.0.0 (serverInfo), protocol 2025-03-26
Capabilities: tools (listChanged: true)
Tools:        6 measured, all annotated readOnlyHint: true
Registry:     finance.haven/financing 1.0.0, remote declared as the same URL
Probe:        initialize, tools/list and four tools/call requests in one session

What the server answered

initialize returned 200 with no credential. Its instruction block draws the boundary before any tool is called: the tools quote published rates, run illustrative simulations and build a prefilled link to Haven's simulator, and they cannot access brokerage accounts, view holdings or balances, open accounts, or execute trades. Applications happen through Haven's own flow, which is where the safety of a read-only financing surface comes from.

tools/list returned six, every one annotated readOnlyHint: true:

get_current_rates          published fixed and variable rates with the all-in figure
simulate_financing         cost for a collateral value, a draw and a term
compare_existing_debt      an existing loan priced against each Haven option
create_onboarding_link     a prefilled simulator link, not an application
search                     reference documents: rates, fees, collateral, savings method
fetch                      full text of a document found with search

Note the parameter names, because they are not the obvious ones: simulate_financing requires collateralValue and drawAmount rather than portfolioValue and amount, and compare_existing_debt requires balance and currentRatePercent. A call with the intuitive names comes back as a schema error naming the field it wanted, which is a fine error but a wasted round trip.

Driven: 200,000 USD against a 1,000,000 USD portfolio

get_current_rates returned asOf: 2026-11-03T20:50:09.001Z with a 0.45 percent Haven fee and four fixed terms:

term   financing   all-in    after-tax (illustrative)
3M     4.5703 %    5.0203 %  3.3454 %
6M     4.6720 %    5.1220 %  3.4199 %
12M    4.8501 %    5.3001 %  3.5503 %
2Y     4.8969 %    5.3469 %  3.5846 %
monthly variable  4.2798 %    4.7298 %  3.1328 %

simulate_financing with a 1,000,000 collateral value, a 200,000 draw and the 12M term returned annual financingInterest: 9,700.20, havenFee: 900, total: 10,600.20, against comparisonAnnualInterest: 17,000 at 8.5 percent, so annualSavings: 6,399.80 and havenIsCheaper: true. The same response carries the collateral arithmetic: LTV 20 percent in the Low band, a facility cap of 70 percent of collateral, a maximum draw of 700,000, a minimum draw of 50,000, and collateralRequiredForDraw: 285,715.

The assumptions are printed rather than implied: the collateral value is the figure the caller supplied, Haven has not verified it or the holdings, the portfolio is assumed to be eligible marginable securities at Interactive Brokers or Charles Schwab, and brokerage requirements can be stricter for concentrated or volatile positions. A fall in collateral value raises LTV and can lead to a margin call or liquidation - the risk that turns this from a rate comparison into a leverage decision is stated in the tool output, not buried in a page.

compare_existing_debt on a 200,000 margin balance at 8.5 percent returned every option at once: the 3-month term saves 6,959.40 a year, the 6-month 6,756, the 12-month 6,399.80, the 2-year 6,306.20, and the monthly variable 7,540.40 - the cheapest line in the table, and the only one with no rate lock. search for "rate lock" returned six documents with stable ids and public haven.finance URLs, which fetch turns into full text.

The rate sheet is nine days old, and the vendor page disagrees

This is the finding worth carrying forward. The MCP's asOf is 2026-11-03T20:50:09Z, nine days before this probe, and its rate table is lower than the one embedded in the vendor's own portfolio-financing page on 2026-11-12:

term   MCP (asOf 3 Nov)   vendor page table (12 Nov)
3M     4.5703 %           4.6169 %
6M     4.6720 %           4.7132 %
12M    4.8501 %           4.9409 %
2Y     4.8969 %           5.0782 %
var    4.2798 %           4.6000 %

Both surfaces describe the same instrument, so one of them is behind. The honest reading is that the MCP serves a published snapshot rather than a live quote, and the asOf field is the tell: any answer that quotes a rate should carry that timestamp, and a borrower comparing against a lender's live quote should ask for the current page. A rate that is 5 to 18 basis points stale does not change a strategy, but it does change a comparison, and the response gives you the date to say so.

Where it sits

It enters beside Milemarker MCP and X1 Wealth MCP, which answer questions about what a household owns; this one answers what that ownership can be borrowed against, and it is the first lending entry here.

Does the server touch a brokerage account?

No. The instruction block states it cannot access brokerage accounts, view holdings or balances, open accounts, or execute trades, and all six tools are annotated read-only. Collateral values are numbers the caller supplies and are recorded as unverified in the response.

What is the smallest draw?

50,000 USD, with the facility capped at 70 percent of collateral value less any Haven financing already outstanding. Both limits appear in the simulation response rather than in documentation.

Can it tell me whether to borrow at all?

It compares cost, not suitability. compare_existing_debt answers which Haven term is cheaper than an existing loan and by how much a year; the leverage decision, the margin-call risk and the tax treatment stay with the person and their adviser.

Integration with CorpusIQ

CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.

This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.

See what CorpusIQ connects

AI answers on your data. See where every number comes from.

Connect your tools and ask in ChatGPT, Claude, or Perplexity. Enterprise-grade answers in minutes. From $29.95/month.

Start free, no credit card
Contact Terms Privacy Pricing