ITM Platform

ITM Platform MCP Server

Connect ITM Platform to AI assistants through the Model Context Protocol. The ITM Platform MCP server lets MCP-compatible clients search projects, inspect budgets, summarize portfolio health, create tasks, log risks and issues, and update project details using your ITM Platform permissions.

It works with Claude, VS Code, Cursor, OpenAI Codex, Windsurf, JetBrains AI Assistant, and any other client that supports MCP.

• Public docs: developers.itmplatform.com/mcp
• npm package: @itm-platform/mcp-server
• Hosted MCP URL: https://api.itmplatform.com/v2/_/mcp/

Quick Start

Hosted connection with OAuth

Use the hosted server if your AI client supports remote MCP servers. There is nothing to install: add the URL, sign in with your ITM Platform account, and approve the requested access.

claude mcp add --scope user --transport http itm-platform https://api.itmplatform.com/v2/_/mcp/

For other MCP clients, use this remote URL:

https://api.itmplatform.com/v2/_/mcp/

OAuth is the recommended setup for most users because your AI client never sees your ITM Platform password or API key.

After adding the server, open your AI client, type /mcp where slash commands are supported, select itm-platform, and complete the ITM Platform OAuth login when prompted.

Local connection with an API key

Use the npm package if you prefer to run the server locally, work behind a firewall, or need to connect to a self-hosted ITM Platform instance.

npx @itm-platform/mcp-server

Your MCP client must pass these environment variables to the server:

Example stdio configuration:

{  "mcpServers": {    "itm-platform": {      "command": "npx",      "args": ["@itm-platform/mcp-server"],      "env": {        "ITM_API_URL": "https://api.itmplatform.com",        "ITM_COMPANY": "{your-account}",        "ITM_API_KEY": "your-api-key"      }    }  }}

To create an API key, log in to ITM Platform, open My Profile, and generate a key from the API Key section.

After configuring the local server, restart your AI client and use /mcp or the client's MCP server list to confirm itm-platform is connected.

What Can an Agent Do?

From simple lookups to fully automated cross-system workflows, MCP unlocks progressively more powerful use cases.

Quick lookup -- Ask a question, get an answer:

"What risks are open across my portfolio?"

Multi-step analysis -- The agent chains multiple tools and synthesizes results:

"Review every project ending this quarter. Flag any with budget overruns, open high-impact risks, or task completion below 60%."

Automated bulk actions -- The agent reads, decides, and writes across projects:

"For every project still in Planning status with a start date in the past, update the status to Execution and create a kick-off checklist task assigned to the project manager."

Scheduled intelligence -- An agent runs on a schedule with no human prompt, pulling overdue tasks every Monday and posting a summary to Slack grouped by project manager.

Cross-system orchestration -- Combine ITM Platform's MCP with other MCP servers (GitHub, Slack, Google Calendar, email). When a developer merges a PR, an agent finds the matching ITM Platform task, marks it complete, and if the project hits 100%, drafts a closure summary and emails the program manager.

The MCP server authenticates as you, calls ITM Platform APIs, and returns only the data your ITM Platform account is allowed to access.

Capabilities

The server exposes 47 MCP tools, 6 resources, and 4 prompt templates.

Read Tools

Write Tools

Write operations confirm the saved state from the ITM Platform REST API. DataMart-backed search results may take up to 60 seconds to reflect recent writes. Validation failures include the actionable message returned by REST instead of only the HTTP status.

Data source routing

Reads come from DataMart whenever DataMart holds the data (quota-free, portfolio-wide); REST is reserved for data DataMart lacks, single-item authoritative reads (get_task), write readbacks, and reference data. Writes always go to REST. When DataMart gains a dataset, the matching GET tools must migrate to DataMart. Full rule and current routing map: zz_Specifications/progress-history-reads-from-datamart.md.

When the account defines custom fields, each session is enriched with per-account context: the server lists the DataMart customFields keys actually in use in the MCP initialize instructions and in the query_datamart tool description, so agents can read and filter custom field values without prior discovery.

Resources and Prompts

Resources give AI clients read-only context such as DataMart schemas and project calendars. Prompt templates provide guided workflows for common analysis tasks:

Authentication and Permissions

The MCP server uses the same identity and permission model as ITM Platform.

OAuth sessions use scopes:

API key sessions use the full permissions of the ITM Platform user who generated the key.

License access:

Your AI assistant does not receive your ITM Platform password or API key. Project data is returned to the AI client you choose, so the AI provider's data-handling policy applies to any data it processes.

Client Setup

Use the public docs for client-specific setup:

• Claude Code, Claude Desktop, VS Code, Cursor, Codex, Windsurf, and JetBrains
• Connect with OAuth
• Connect with an API key
• Troubleshooting

For any MCP-compatible client, the two connection values are:

After adding either connection, open the client's MCP command or server list. In clients that support slash commands, type /mcp, select itm-platform, and authenticate when prompted.

Self-Hosting

For a local stdio server, configure ITM_API_URL, ITM_COMPANY, and either ITM_API_KEY or ITM_TOKEN.

For an HTTP server with OAuth, configure:

When deployed behind a reverse proxy, ITM_AUTH_URL can point to a server-to-server address while ITM_AUTH_PUBLIC_URL must be reachable by AI clients.

Development

Requirements:

• Node.js 20 or later
• npm

Install dependencies, run tests, and build:

npm installnpm testnpm run build

Run the HTTP development server:

cp .env.sample .envnpm run dev

The package entry point is dist/server.js; the npm executable is mcp-server.

Troubleshooting

If tools do not appear in your AI client, confirm that the server configuration is in the correct file for that client, restart the client, and check that npx @itm-platform/mcp-server runs successfully for local setups.

If authentication fails, regenerate your API key or reconnect the OAuth server so your client receives a fresh token.

OAuth sessions automatically retry once on a downstream 401 by re-exchanging the OAuth bearer token for a fresh session token. This handles cases where the session token is invalidated externally (e.g. by a concurrent browser login). If 401 errors persist, the OAuth bearer token itself has likely expired and the AI client needs to re-authenticate.

If a write succeeds but a later search shows old data, wait up to 60 seconds. Writes are confirmed from the REST API immediately, while DataMart search indexes update asynchronously.

More Help

• MCP docs: modelcontextprotocol.io
• ITM Platform help: helpcenter.itmplatform.com
• ITM Platform developer docs: developers.itmplatform.com