Anaplan MCP

Unofficial MCP server for Anaplan

Setup guide: platform-aware install walkthrough for Windows, macOS, and Linux with OAuth2 / Certificate / Basic Auth options.

A Model Context Protocol (MCP) server that connects AI assistants to Anaplan's Integration API v2. Gives LLMs like Claude direct access to browse workspaces, manage data, run imports/exports, and administer models through 70 structured tools, using your existing Anaplan credentials and permissions.

Built in TypeScript. Supports both stdio (local) and Streamable HTTP (remote) transports. Works with Claude Desktop, Claude Code, claude.ai, and any MCP-compatible client. Includes a built-in orchestration guide that teaches the AI assistant the correct tool sequences for every workflow.

Why This Exists

Anaplan's Integration API is powerful but requires technical expertise to use directly. Most teams rely on a handful of model builders to navigate complex models, extract data, and run imports - creating bottlenecks when others need access to the same information.

This server wraps the API in 70 structured tools that AI assistants like Claude can call on your behalf. Explore models, pull data, run actions, and onboard new team members - all by asking in plain English instead of writing API calls or waiting for someone who knows the model.

For business users: Stop waiting for someone to pull data or explain how a model works. Ask Claude to show you the numbers, walk you through module structure, or run your regular imports.

For model builders and consultants: Analyze model structure, trace formula dependencies, review line item configurations, and identify performance issues - all through conversation instead of clicking through hundreds of modules manually.

For IT and platform teams: Standard API access using your existing authentication and permissions. No new credentials, no elevated access. Open source for auditability. Anaplan data is processed by your AI assistant - locally or through your provider's environment depending on your setup.

Related MCP server: Equinix Fabric MCP

Common Use Cases

Model Documentation:

• "Show me the structure of the Supply Planning model"

• "List all line items and their formulas in the Revenue module"

• "What dimensions does the Margin line item use?"

Data Review:

• "Pull the current pricing data for all products"

• "Show me which list items were recently added"

• "Read the forecast numbers for Q3 and summarize"

Impact Analysis:

• "What modules use the Product list as a dimension?"

• "Show all line items that reference Cost Per Unit"

• "Which views include the Region dimension?"

Automation:

• "Run the monthly demand import and show me the result"

• "Export sales actuals and save to Downloads"

• "Add these 50 new products to the master list"

Onboarding:

• "Walk me through the modules in this model"

• "How is this model structured? What are the key lists?"

• "What changed in this model since last quarter?"

What It Can and Can't Do

What it can do

• Browse workspaces, models, modules, lists, and views

• Read cell data and list items

• Write cell values and manage list items (add, update, delete)

• Run imports, exports, processes, and delete actions

• Upload and download files

• Manage models (open, close, delete, set periods and fiscal year)

• Query users, versions, and task history

Model Building Limitations

The Anaplan API does not support:

• Creating modules or line items programmatically

• Defining formulas through API

• Building model structure from scratch

• Configuring model calendar programmatically

For model building, use Anaplan's UI or Agent Studio.

Prerequisites

• Node.js 18+ - download here

• An Anaplan account with API access (any auth method - basic, certificate, or OAuth)

• An MCP-compatible client - Claude Desktop (recommended), Claude Code, or any other MCP client

Setup

1. Clone and build

git clone https://github.com/larasrinath/anaplan-mcp.git
cd anaplan-mcp
npm install
npm run build

2. Connect to Claude Desktop

Claude Desktop is the easiest way to use this server. Here's how to set it up:

Step 1: Open the config file

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

• Windows: C:\Users\<YourUsername>\AppData\Roaming\Claude\claude_desktop_config.json

If the file doesn't exist yet, create it with {} as the contents.

Tip (Windows): You can type %APPDATA%\Claude in the File Explorer address bar to jump straight to the folder.

Step 2: Add the Anaplan server

Replace <path> with the absolute path to your cloned repo (e.g. /Users/you/anaplan-mcp on macOS/Linux or C:/Users/you/anaplan-mcp on Windows - always use forward slashes).

Choose one auth method only. For most users, use OAuth2 so Claude can show a sign-in link in chat. Do not set OAuth, certificate, and basic env vars together.

Recommended: OAuth2 (device grant)

{
  "mcpServers": {
    "anaplan": {
      "command": "node",
      "args": ["<path>/dist/index.js"],
      "env": {
        "ANAPLAN_CLIENT_ID": "your-client-id"
      }
    }
  }
}

On first use, Claude shows a link in chat, approve it in Anaplan, then retry your request. OAuth tokens are kept in memory only. If the MCP process restarts, or an OAuth session is idle for more than 60 minutes, you'll be prompted to authorize again unless you provide ANAPLAN_REFRESH_TOKEN yourself.

OAuth support is device grant only. ANAPLAN_CLIENT_SECRET, ANAPLAN_OAUTH_AUTHORIZATION_CODE, and ANAPLAN_OAUTH_REDIRECT_URI are ignored by the server.

If you do not want OAuth, use one of these alternatives instead:

Certificate auth

"env": {
  "ANAPLAN_CERTIFICATE_PATH": "/path/to/cert.pem",
  "ANAPLAN_PRIVATE_KEY_PATH": "/path/to/key.pem"
}

ANAPLAN_CERTIFICATE_ENCODED_DATA_FORMAT can be added optionally; defaults to v2. Set v1 only for legacy tenants.

Basic auth

"env": {
  "ANAPLAN_USERNAME": "user@company.com",
  "ANAPLAN_PASSWORD": "your-password"
}

Use this only if you want email/password auth instead of OAuth or certificate auth. If your Anaplan account uses SSO, basic auth may not work unless your tenant allows exception users.

If your config file already has content, add mcpServers inside the existing top-level object - don't create a second {} block.

Step 3: Restart Claude Desktop

Quit Claude Desktop completely (right-click the system tray icon and quit - don't just close the window) and reopen it. You should see the Anaplan tools available (look for the hammer icon in the chat input area).

Troubleshooting:

• "Unexpected non-whitespace" error - Your JSON is invalid. Make sure there's only one {} object in the file and no trailing commas. Paste your config into jsonlint.com to check.

• Server disconnected - Run node C:/path/to/anaplan-mcp/dist/index.js in a terminal to see the actual error. Common causes: wrong path in args, missing npm run build, or Node.js not installed.

• 401 Unauthorized when using tools - Your Anaplan credentials are wrong, or your account uses SSO (in which case basic auth won't work - use certificate or OAuth2 instead).

• OAuth refresh failed / reauthorization required - The MCP server is up and reached Anaplan, but the saved OAuth session is no longer valid. Re-authorize in your MCP client, then retry the tool.

• Model opens in the browser but is missing from discovery / returns metadata 404 - Discovery can omit accessible models, including models in another customer/tenant. Extract the IDs after /workspaces/ and /models/ in the browser URL and pass them directly to show_modules or show_imports. show_modeldetails now checks these read-only endpoints if global model metadata returns 404 and reports verified access separately from unavailable metadata. Names require discovery, so use explicit IDs for omitted models. This does not grant additional permissions or prove that writes/actions are allowed; no customer prefix or header is added to the Integration API.

Connect to Claude Code

Copy .mcp.json.example to .mcp.json and fill in your credentials. The file is gitignored by default - never commit credentials to version control.

cp .mcp.json.example .mcp.json

Alternatively, use the CLI:

claude mcp add anaplan -- node /absolute/path/to/anaplan-mcp/dist/index.js

Other MCP clients

Any MCP-compatible client that supports stdio transport can connect. The server expects to be launched as a subprocess with stdin/stdout for communication. Pass Anaplan credentials via environment variables.

Browser-based AI (claude.ai, ChatGPT)

The server also supports Streamable HTTP transport for remote MCP connections from claude.ai, ChatGPT, and other browser-based AI assistants. Deploy to a cloud platform (Fly.io recommended) and connect via the remote MCP integration settings.

Remote HTTP mode is designed for per-session Anaplan OAuth, not a single shared Anaplan user. Set ANAPLAN_CLIENT_ID on the server so each remote session can authorize against Anaplan with its own identity. Basic auth (ANAPLAN_USERNAME/ANAPLAN_PASSWORD) and certificate auth (ANAPLAN_CERTIFICATE_PATH/ANAPLAN_PRIVATE_KEY_PATH) are intentionally not supported in remote HTTP mode — they would collapse every session onto one shared Anaplan identity, breaking per-user permissions and auditability. They remain available for stdio/local use only. If you want an extra outer gate in front of the endpoint, you can also set ANAPLAN_MCP_HTTP_AUTH_TOKEN and have your client or reverse proxy send it as Authorization: Bearer <token>.

See the Remote Deployment Guide for full setup instructions, platform recommendations, and troubleshooting.

Configuration

Environment variables

All configuration is done through environment variables. There are no config files, CLI flags, or settings menus.

You only need one set of credentials. If multiple are configured, the server picks the highest-priority method automatically.

HTTP transport security

These apply only to npm run start:http / remote MCP deployments:

Where to set environment variables

• Claude Code config: Use the "env" block in .mcp.json (file is gitignored by default)

• Claude Desktop config: Use the "env" block in the JSON config (keeps credentials scoped to the server)

• Shell profile: Export in .bashrc / .zshrc for Claude Code CLI usage

• System environment: Set at the OS level if you prefer

Security note: Never commit credentials to version control. Env files and MCP config files are gitignored by default in this repo.

Permissions and Safety

What the server can do

This server has full access to whatever your Anaplan credentials allow. The 70 tools cover both read and write operations:

• Read-only tools (safe to use freely): show_* tools, read_cells, get_list_items, download_file, get_action_status

• Write tools (modify data): write_cells, add_list_items, update_list_items, delete_list_items

• Action tools (trigger Anaplan processes): run_import, run_export, run_process, run_delete

• Admin tools (model management): close_model, open_model, bulk_delete_models, set_currentperiod, set_fiscalyear

Tool approval in Claude Desktop

Claude Desktop prompts you before each tool call. You'll see the tool name and parameters, and can approve or deny. This gives you a chance to review before any action runs. You can also use the "Allow for this chat" option for tools you trust.

Recommendations

• Start with read-only. Ask Claude to explore your workspaces and models before running any write operations. Get comfortable with the tool output first.

• Test in a dev workspace. If you have a non-production Anaplan workspace, use that while getting familiar with the tools.

• Use least-privilege credentials. If your Anaplan admin can create a service account with limited workspace access, use that instead of your personal admin account.

• Review before confirming write operations. When Claude proposes to run an import, write cells, or delete items, read the parameters carefully before approving.

• Exports and imports are asynchronous. The server polls until they complete (up to 5 minutes). You can cancel a running task with cancel_task if needed.

Tools

Model Exploration (37 tools)

Bulk Data Operations (28 tools)

Transactional Operations (5 tools)

Orchestration Guide

The server exposes a built-in MCP resource (anaplan://orchestration-guide) that AI assistants read automatically. This guide teaches the correct tool sequences for every workflow category:

• Navigation -- workspace -> model -> module -> line items/views

• Reading data -- discover modules, find views, read cells (with fallback to large volume reads for >1M cells)

• Writing data -- resolve line item dimensions and item IDs before calling write_cells

• Bulk imports -- inspect import definition, upload data to source file, run import, check status, download error dump if failed

• Bulk exports -- single-step run_export handles the full lifecycle

• Processes -- run chained actions, monitor with get_action_status

• Large volume reads -- create request, poll until complete, download pages, clean up

• List mutations -- find list items, then add/update/delete

Every tool description also includes prerequisite hints ("Use show_imports first to find importId") and parameter descriptions explain where each value comes from ("from show_lineitems or show_alllineitems"). Key workflow tools append "Next steps" guidance to their responses.

Architecture

src/
  auth/       # Authentication providers (basic, certificate, oauth) + token manager
  api/        # HTTP client with retry logic + 17 domain-specific API wrappers
  tools/      # MCP tool registrations (exploration, bulk, transactional) + response hints
  resources/  # MCP resource content (orchestration guide)
  server.ts   # Wires auth > client > APIs > MCP server + registers resources
  index.ts    # Entry point (stdio transport)
  http.ts     # Entry point (Streamable HTTP transport)

docs/
  README.md   # Integration documentation index
  api/        # Anaplan Integration API v2 reference
  architecture/ # Runtime diagrams (request flow, trust boundary, subsystem map)
  guides/     # Integration tool workflows and remote deployment

Three layers:

1. Auth layer - pluggable providers behind a common AuthProvider interface. The AuthManager selects the right provider from env vars and handles token lifecycle.

2. API layer - AnaplanClient handles all HTTP communication with the Anaplan API. 17 domain wrappers provide typed methods for each endpoint. Auto-paginates list endpoints using Anaplan's meta.paging metadata.

3. Tools layer - registers MCP tools on the server with zod schemas for input validation. Each tool delegates to the appropriate API wrapper and formats results. Key tools include next-step hints to guide multi-tool workflows.

See the documentation index for Integration API and tool guides. For detailed runtime diagrams (request flow, trust boundary, subsystem map) see docs/architecture/overview.md.

Custom Skills

The skills/ folder holds Claude Code project-level skills - reusable instruction sets that the AI assistant loads automatically during your session.

A template is provided at skills/example.md. Copy it, rename it, and fill in your instructions. Claude Code will pick it up as a skill available in this project.

Skills are gitignored by default (personal workflows vary), with only the example tracked. Add your own without worrying about committing them.

Disclaimers

Unofficial personal project - not affiliated with, endorsed by, or supported by Anaplan. Uses the official Anaplan Integration API v2 - no undocumented endpoints. Users are responsible for compliance with Anaplan's Terms of Service. No warranty provided; use at your own risk.

License

MIT - see LICENSE file for details. Covers the code in this repository only. Anaplan's API and service are subject to Anaplan's Terms of Service and Acceptable Use Policy.