Local MCP Server for Resource Planning

Local MCP Server for Resource Planning

A connected AI assistant can use the local MCP server in Rillsoft Project 10 to examine open roles, suitable employees and capacity gaps in the open plan. You review its proposal and decide which assignments to make; Rillsoft Project calculates dates, demand and workload. This gives resource managers in engineering and industrial teams a way to discuss staffing against the computed schedule.

The MCP server provides another route to the scheduling engine alongside the user interface. It runs inside the visible Windows application and is reachable only at 127.0.0.1. The server is local; what project data your AI client sends to its provider depends on that client’s settings and privacy terms.

Try Rillsoft Project for 30 days · Set up the connection on rillsoft.ai

What the planner does with it

A connected assistant such as Claude or Codex answers questions straight from the computed schedule:

• Who is overloaded next week? – the warning lists for delayed, overloaded and unassigned tasks, across projects too.
• Where is the bottleneck per role, team or machine? – demand against supply in the time grid of the capacity view, with the shortfall as the answer.
• Who could take this task? – candidates with availability, free capacity, cost and schedule conflict, before a person is named.
• How far does the plan deviate from the baseline? – comparison for time, effort and cost, progress per task and status date.

Beyond that, a client builds and changes project structures: create subprojects, tasks, dependencies, roles, people, machines and material; restructure, split, schedule. In total the server covers ten tool areas – from reading the project through capacity analysis, staffing and site planning to baseline, actual data, resource pool, portfolio and presentation.

What stays the same

• Rillsoft Project computes, the AI works with the result. The scheduling engine remains the single source of truth for dates, durations, workload and the critical path. The tools read the result – they do not compute it.
• Every change is one undo step. Every mutating call is exactly one step that Ctrl+Z takes back. Several mutating calls create several undo steps; undo does not replace checking the resulting plan.
• Read-only mode. In read-only mode the server offers reading tools only; mutating calls are rejected with a clear error.
• Data sovereignty. The server binds to 127.0.0.1 only. What your AI client transmits to its provider is governed by that provider’s privacy policy – not by Rillsoft Project.

An evidenced example: staffing the packaging cell

The synthetic “Packaging cell P-80” example starts with 26 tasks carrying open roles. A documented run on build 10.0.624 dated 29 September 2026 produced 34 employee assignments across 24 tasks. Four tasks retained open or partly filled employee roles; the subsequent check reported no employee overload. This is the result of that example, not a guarantee for other plans.

The workflow separates candidate search, a reasoned proposal, your confirmation and the assignment itself. The instruction tells the assistant to wait for confirmation; this is not a universal server-side approval dialog before every write. Read the evidence and staffing workflow.

Read the practical guide to proposals, confirmation and remaining gaps: AI Agents for Project Resource Allocation.

Requirements

• Rillsoft Project 10 on Windows; earlier versions have no MCP server.
• An MCP client that speaks the Model Context Protocol over Streamable HTTP (revision 2025-06-18). Claude and Codex are verified.
• The relevant licensed capability for the candidate list and staffing assistants. Connecting a client does not unlock licensed functions. Contact us to confirm the edition needed for your workflow.
• An open project or portfolio with maintained roles, calendars and resources. Conflicts outside that open scope are not covered.
• Write access to apply assignments; read-only mode is enough to inspect the plan.
• Optionally an API key in the settings – advisable on multi-user and terminal-server machines, because 127.0.0.1 applies machine-wide.

An MCP call carries no separate user identity. Review the client and its data transfers before using your own employee data. Start with the synthetic examples.

Without AI: scripts and integrations

The same endpoint can be used without an AI assistant. A script initialises the session, reads the tool contract and calls the same tools – for instance to fetch the capacity report every morning or to feed in actual data. Bundled industry examples with synthetic data allow a first test without a project of your own.

Working with the Integration Server

The MCP server works on the project that is open in the desktop client – whether it comes from a local file or from Rillsoft Integration Server. Projects, documents, portfolios and folders on the server are thus reachable via MCP through the open client; multi-user operation, permissions and central storage remain the server’s job.

Check SAP orders for shortfalls via MCP

Production and maintenance orders from SAP S/4HANA can be checked for capacity shortfalls in Rillsoft Project without changing SAP. An AI client such as Claude connects two MCP servers for this: the MCP Gateway of SAP Integration Suite reads orders, operations and work center capacities; the local MCP server of Rillsoft Project 10 creates the project from them. Rillsoft Project calculates dates and capacity, the AI only transfers the data. Nothing is written back to SAP, and MCP offers no automatic capacity levelling with date shifts.

The test bench documented on rillsoft.ai uses SAP sandbox data: a battery production with 6 orders and 48 operations and a maintenance case with 30 orders and 117 operations. The complete route can be reproduced free of charge with the SAP BTP trial and the 30-day trial of Rillsoft Project. Guide, data flow, limits and both examples: Connect Rillsoft Project to SAP.

Continue on rillsoft.ai

Guides, prompts and developer documentation live on the AI hub rillsoft.ai:

• Get started – enable the server, connect a client, first safe call.
• Prompts – proven questions for the open project, from delayed tasks to projects competing for resources.
• Questions from management – 22 management questions on status, schedule, capacity and costs; one prompt per verified question, together a management dashboard.
• Developers – tool contract, schemas and error codes for scripts and your own integrations.
• SAP connection – read SAP orders through the MCP Gateway of Integration Suite and check them for capacity shortfalls in Rillsoft Project; setup, two examples and a complete reproduction guide.

Try Rillsoft Project 10

The MCP server is part of Rillsoft Project 10: try it free for 30 days. The other new features of this version at a glance: New in Rillsoft Project 10.0.

Editorial responsibility: Rillsoft. Content reviewed: 31 October 2026; example evidence: 29 September 2026, build 10.0.624; SAP evidence: 24 October 2026, build 10.0.626.