MCP server

Connect DeepCell to
any MCP client.

One MCP tool exposes the whole DeepCell CLI to web platforms like claude.ai, Manus, and ChatGPT — the same 20+ commands, over OAuth 2.1 with PKCE.

MCP server

MCP server

One MCP tool exposes the entire DeepCell CLI to web-based AI agents. Enterprise-grade OAuth 2.1 security with PKCE.

Claude.ai Settings

Add this URL as a remote MCP server in Claude.ai settings:

https://your-domain.com/mcp

Go to Claude.ai → Settings → Integrations → Add MCP Server → paste the URL above. OAuth authentication is handled automatically.

Start a chat in your platform

Use deepcell tool to create a dcf model

Works with

claude.ai

Anthropic's web-based AI assistant with native MCP support.

Manus

Autonomous AI agent platform with MCP tool integration.

Any MCP client

Any platform supporting the Model Context Protocol standard.

Running a local agent with a shell?

Claude Code, Cowork, OpenWork, or your own terminal don't need MCP — they run the DeepCell CLI directly, which is faster and unlocks sign-in, the Excel and PowerPoint exports, and the sync commands that MCP blocks.

What your agent gets

One MCP tool carries the whole CLI — minus the handful of commands MCP blocks on purpose: sign-in, the Excel and PowerPoint exports, and the git sync commands.

Agent-ready output

Concise plain text by default — token-light for agents. Pass --format json when a tool needs structured output.

Version control

Git-like operations — log, diff, restore — for structured financial data.

Batch editing

Edit single values or pass a JSON file for bulk updates across models.

Reasoning graph

Read and write the claims, assumptions, and evidence behind a number — not just the number.

Dimensional queries

Address any value by Item, Context, and Status — not row and column.

OAuth 2.1 + PKCE

Industry-standard authorization. No client secrets required for web agents.

Token rotation

15-minute JWT lifetime with refresh tokens rotated on every use.

Scoped access

Blocked commands prevent auth-bypass. Every MCP call is logged for audit.

What it looks like when your agent runs it

Commands return concise plain text by default — token-light, easy for any agent to parse and act on.

terminal
# Query a projected revenue value $ deepcell query model.deepcell Revenue FY2025E projected 1500000 # Batch edit from a JSON file $ deepcell edit model.deepcell --batch changes.json 3 edit(s) applied rev:a1b2c3d4 ✓ 3 edit(s) applied # Read back every version of the model $ deepcell log model.deepcell a1b2c3d4 2026-07-28 Raise FY2026 growth to 18% 9f8e7d6c 2026-07-21 Import Q2 actuals

How it works

Step 01

Connect

Paste the MCP URL into your AI platform, or copy the config below — it already has your API key in it. One config line either way.

Step 02

Authenticate

OAuth 2.1 + PKCE — you sign in securely in the browser, with no secrets stored client-side. Or use an API key you can revoke at any time.

Step 03

Model

Ask your AI to build financial models, query data, or trace the reasoning behind a number. 20+ commands through one tool.

Get started

One tool. Any MCP client.
Your agent handles the rest.

Paste one config and your agent gets 20+ commands for querying, editing, and reasoning over financial models.