Calling Forensic Schedule Tools from Cline, Cursor, and Any MCP Client

AI coding assistants got tool use last year. Most of them now ship with file editing, web search, and terminal access wired in by default. There is — until now — almost nothing on the construction-scheduling side of the toolbox. If your AI assistant has never opened a Primavera P6 XER, that's because nobody has put forensic-scheduling tools inside the protocol it speaks.

Critical Path Partners' forensic-scheduling engine is hosted at mcp.criticalpathpartners.ca and callable from any MCP-compatible AI client today — Cline, Cursor, Claude Desktop, Claude in the browser, Continue, Zed, and anything else that supports the Model Context Protocol. Thirteen tools. One URL. Setup is a single config-file edit. Below is a real run from Cline, captured this week.

The Demo

Cline (open-source coding agent, running Anthropic Claude Sonnet 4.5 via the user's Max subscription) was given a single instruction: call xer_parser on sample.xer and tell me the project name, data date, task count, and relationship count.

What actually happened: Cline read the file contents from the workspace, sent them to the hosted MCP at mcp.criticalpathpartners.ca/mcp, the server parsed the XER and returned a structured JSON with the table summary, and Cline narrated four fields from that JSON in plain English. Total elapsed time: about ten seconds. Total per-token cost: $0.00 — the LLM ran on the user's Max subscription via Cline's Claude Code provider, and the MCP server is free.

This is not a chatbot interpreting an XER. The MCP server runs the actual parsing code — the same code that powers every CPP forensic deliverable. The AI orchestrates. The methodology executes. The numbers come back grounded.

Setup in 30 Seconds

Every MCP client has a config file. The entry is the same shape every time — a name, a transport type, and the URL.

Cline (cline_mcp_settings.json)

{
  "mcpServers": {
    "cpp-cpm-engine": {
      "type": "streamableHttp",
      "url": "https://mcp.criticalpathpartners.ca/mcp"
    }
  }
}

Cursor (~/.cursor/mcp.json)

{
  "mcpServers": {
    "cpp-cpm-engine": {
      "url": "https://mcp.criticalpathpartners.ca/mcp"
    }
  }
}

VS Code Copilot Agent (.vscode/mcp.json)

{
  "servers": {
    "cpp-cpm-engine": {
      "type": "http",
      "url": "https://mcp.criticalpathpartners.ca/mcp"
    }
  }
}

Claude Desktop

The new Claude Desktop wires custom MCP servers through the UI rather than the JSON config: Settings → Connectors → Add custom connector, paste https://mcp.criticalpathpartners.ca/mcp, save. The tool list populates automatically.

What You Get

Thirteen forensic tools, each tied to an AACE recommended practice or DCMA assessment criterion. Every output is a structured JSON return that an AI can summarize, a downstream tool can ingest, or a human can audit.

Why This Matters

Construction lawyers, project managers, and claims consultants are using AI assistants as their default research interface now. When the lawyer asks the AI to "run a DCMA-14 health check on this XER and tell me which schedule looks defensible," the tools that are wired into MCP get called. The tools that exist as a PDF whitepaper on a consulting firm's website do not.

Until this week, the answer to that question was no AI assistant can do this — you'd have to license SmartPM or hire a forensic scheduler to manually run the analysis. Now the answer is one URL in your config, results in fifteen seconds, structured JSON you can audit. The expert opinion still requires the expert. The rote work that gets the expert to the opinion does not.

What This Doesn't Replace

Same line as every other AI-in-construction piece on this blog: the methodology executes, the AI orchestrates, the human decides. dcma14_health_check can tell you that 31% of activities lack predecessors. It cannot tell you whether the contractor's narrative about why is credible. forensic_windows_analysis can attribute critical-path drift to specific activity codes within a window. It cannot tell a hearing panel which party should bear that drift under the contract.

The MCP server is the rote-work layer — XER parsing, metric computation, AACE-canonical attribution math, audit-trail generation. The forensic analyst's judgment, expert qualification, courtroom credibility, and contract interpretation sit on top of that layer. The faster the rote work runs, the more time the analyst has for the parts that matter.