For AI-native developers · MCP / REST · day one

A financial modeling MCP
your agent can call.

Typed IR. DAG. Sensitivity. Scenarios. Excel export. Nine tools. One npx command, and your agent harness has a deterministic finance backend behind it. No custom-tool glue, no JSON pretending to be a model.

See the tool surface ↓Get an API key →

agent builders · live pulse0 compiles · 0 in last window

MCP+ REST

Both protocols, day one. Your agent calls the same engine your backend can curl.

quickstart →

9tools

flatland_init, build, compile, scenario, sensitivity, save, load, list, export. The full surface.

tool catalog ↓ →

87ns

Per-record sensitivity propagation. Compiled cold to warm in under a second.

bench · sensitivity →

bit-identical

Same IR always produces same outputs. Determinism is a guarantee, not a marketing claim.

determinism →

50/ month free

Then $0.10 per answer. No seat licenses. Sensitivity sweeps count as one regardless of fan-out.

pricing →

1command

npx flatland-setup auto-detects Claude Code and Cursor. Writes config, runs end-to-end test.

install →

Specimen · MCP tool catalogyour agent sees these · billable shown in signal-orange

Tool

$

Purpose

Returns

flatland_init

—

Load skills and templates into the agent context. Call first every session.

skills_catalog, version_info, usage_summary

flatland_create_model

—

Create a new model. Set name, currency, period configuration.

model_id, name

flatland_build_model

—

Add typed drivers to a model. Accepts batched driver definitions.

model_id, drivers[]

flatland_compile

●

Topologically sort the DAG, evaluate, run dimensional inference, run assertions. The headline tool.

bindings, assertions, latency_ms, bit_identical

flatland_compile_scenario

●

Compile a sparse overlay against the base model. Returns diff attribution.

bindings, diff_vs_base, latency_ms

flatland_sensitivity

●

Perturb one or more drivers, walk the DAG, return elasticity-ranked output deltas.

ranked_drivers, elasticities, latency_ms

flatland_save_model

—

Persist the current model to local storage (~/.flatland/models/).

model_id, path

flatland_load_model

—

Load a previously saved model by name.

model_id, drivers, assertions

flatland_export_excel

—

Multi-sheet workbook with live formulas via named ranges. Rubric-scored.

path, rubric_score, sheet_count

Three tools are billable: flatland_compile, flatland_compile_scenario, flatland_sensitivity. Each counts as one “answer.” A sensitivity sweep across 50 drivers is one answer, not 50. Everything else (listing, loading, exporting, inspecting) is free.

Your agent calls.
A real compiler answers.

flatland → /workshopflatlandfi.com/m/________

Intake

What are you modeling?

What stage?

What vertical?

12 drivers · 4 assertions · ready

Your compile contributes to the live system heartbeat. Anonymized: no model contents, names, or assumptions leave your browser.

// Answer the three questions on the left. Your IR materializes here.

Dependency graph

Output · forward 24 months

ARR$3.2M

Gross margin78.4%

Runway at current burn19 months

Burn multiple1.12 ✓

Compile1.84s · bit-identical

Artifacts

Download .xlsx32.4 KB · 100/100 Shareable URLcopy →View IR JSONapplication/json

Ask anything

v2 · Phase 1.5

Specimen · one tool call, end to endMCP · tools/call · flatland_compile

RequestPOST /mcp · tools/call

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "flatland_compile",
    "arguments": {
      "model_id": "ndo_q3"
    }
  }
}

Header: X-API-Key: fl_live_…

Response (200)3 / 4 assertions pass

{
  "jsonrpc": "2.0",
  "id": 1,
  "result": {
    "model_id": "ndo_q3",
    "bindings": {
      "arr": 1617720,
      "gross_margin": 0.82,
      "runway_months": 18.27,
      "burn_multiple": 3.86
    },
    "assertions": [
      { "id": "gm_floor",          "passed": true },
      { "id": "runway_target",     "passed": true },
      { "id": "burn_mult_ceiling", "passed": false },
      { "id": "arr_floor",         "passed": true }
    ],
    "latency_ms": 47,
    "bit_identical": true
  }
}

One billable answer. The model produced a number andtold your agent which assumption broke. That's the difference between an answer and a guess.

Nextpick where you want to go deeper

If you're embedding

The platform surface.

Three integration surfaces, multi-tenant isolation by architecture, OEM-friendly. The only Contact-Sales page on the site.

Read /platforms →

If you're a founder

What your end-user sees.

The page the founder hits when their agent compiles their fundraising model. Useful for understanding what your downstream user is doing.

Read /founders →

If you've seen enough

Get the key. Try it.

Two minutes from here to a working compile on your machine. No card.

Get an API key →

Bring your agent.
We'll bring the math.

Flatland · index of everything

Product

For

Comparisons

Disasters

Docs

Quickstart

Discover

Company

Contact · info@