Flatland MCP

Financial reasoning infrastructure for AI agents.
Typed models. Deterministic compilation. Cryptographic receipts.

---

What This Is

Flatland is a financial modeling engine that connects to AI agents via MCP. Your agent describes a business, calls Flatland's tools, and gets back a compiled financial model — typed, verifiable, and reproducible.

The agent brings language. Flatland brings structure.

┌─────────────────────────────────────────────────────────────────┐
│  Your AI agent  (Claude Code / Cursor / Codex / any harness)    │
│  ↓  MCP tools                                                   │
│  Flatland Engine  (api.flatlandfi.com)                          │
│  ↓  typed IR → compiled outputs + signed receipt                │
│  Your machine  (client-owned receipt chain)                     │
└─────────────────────────────────────────────────────────────────┘

Every number is compiled, not narrated. Every output ships with a cryptographic receipt your client holds — Flatland retains nothing server-side.

---

Quick Connect

Claude Code

Add to ~/.claude/claude_desktop_config.json:

{
  "mcpServers": {
    "flatland": {
      "command": "npx",
      "args": ["-y", "flatland-setup"],
      "env": {
        "FLATLAND_API_KEY": "your-key-here"
      }
    }
  }
}

Cursor / Windsurf

Add to .cursor/mcp.json (or equivalent):

{
  "mcpServers": {
    "flatland": {
      "command": "npx",
      "args": ["-y", "flatland-setup"],
      "env": {
        "FLATLAND_API_KEY": "your-key-here"
      }
    }
  }
}

Any MCP-compatible harness

FLATLAND_API_KEY=your-key npx flatland-setup

→ Get your API key at flatlandfi.com

---

Tool Catalog

Model Lifecycle — FREE

Tool	Description
flatland_init	Initialize and load server-side domain skills
flatland_create_model	Create a model (name, grain, period range, currency)
flatland_load_model	Load a saved model from local durable storage
flatland_save_model	Persist the active model to disk
flatland_list_models	List all locally saved models
flatland_explain	Trace the upstream compute chain for any driver
flatland_diff_scenarios	Compare two named scenarios
flatland_list_decisions	Retrieve the decision trace log for a model
flatland_verify	Stateless verifier — verify a signed receipt against inputs (no API key required)

Build — FREE

Tool	Description
flatland_add_driver	Add a typed assumption driver (constant, growth rate, ratio, etc.)
flatland_add_computed	Add a computed driver with a formula
flatland_bulk_add	Add 3+ drivers atomically
flatland_update_driver	Update a driver's value, formula, or metadata
flatland_remove_driver	Remove a driver (cascade to dependents)

Compute — Metered ($0.10/answer, 50 free/month)

Tool	Description
flatland_compile	Compile the active model; returns all computed values
flatland_sensitivity	Rank assumptions by impact on a target driver
flatland_impact_preview	Preview downstream impact of proposed changes
flatland_create_scenario	Create a named scenario with driver overrides

Export

Tool	Description
flatland_export	Export as HTML (institutional audit artifact), Excel, JSON, or Markdown

Total: 37 MCP tools. See full tool reference →

---

Why Flatland, Not a Spreadsheet or a Chat Model

Capability	Flatland	Spreadsheet	LLM-generated model
Typed driver graph (no anonymous cells)	✅	❌	❌
Deterministic, reproducible compilation	✅	partial	❌
Cryptographic receipt chain	✅	❌	❌
Client-owned custody (server retains nothing)	✅	n/a	❌
Stateless, offline-verifiable outputs	✅	❌	❌
Agent-native / MCP-first	✅	❌	❌
Acyclic DAG — no circular dependencies	✅ enforced	❌	❌

The engine never calls an LLM. The LLM calls the engine. Computation is deterministic and independent of which AI model your agent uses — today Claude Code, tomorrow Gemini CLI, same compiled numbers.

---

Architecture

Flatland uses a per-call IR model. The Intermediate Representation (IR) is a typed, declarative JSON-serializable DAG of drivers. It lives on your machine. Each compile call sends the IR to Flatland's engine, receives results plus a signed receipt, and the server discards the request.

flatland_create_model → flatland_add_driver(s) → flatland_compile
                                                        ↓
                                            compiled values + signed receipt
                                                        ↓
                                            your machine (server retains nothing)

Receipt chain (Flatland Institutional): Every compiled output ships with an HMAC-signed receipt that includes a parent_hash binding it to the prior receipt, forming a tamper-evident chain. flatland_verify is free and stateless — pass the receipt and the original IR, the server recomputes and verifies the HMAC. No API key required. Verifiable offline.

---

Pricing

Plan	Compiled answers	Price
Free	50 / month	$0
Metered	Beyond 50	$0.10 / answer
Institutional	Signed receipt chain, audit HTML, decision trace	Contact us

Free includes all model build tools, flatland_verify, flatland_diff_scenarios, flatland_explain, and flatland_list_decisions. Only flatland_compile, flatland_sensitivity, and flatland_impact_preview are metered.

---

Flatland Agent

For a self-contained decision session in your terminal:

npx flatland-agent

Six-stage Structured Decision Session → compiled model → signed HTML export with GO / NO-GO / UNCERTAIN verdict. Bring your own Anthropic or OpenAI key.

---

Links

Resource	URL
Docs & quickstart	flatlandfi.com/docs
Get an API key	flatlandfi.com
Flatland Agent CLI	npx flatland-agent
Status	status.flatlandfi.com

---

License

MIT — see LICENSE.