rillsoft/rillsoft-project-mcp

Deutsch
Rillsoft AI is the AI gateway to Rillsoft Project, the specialist solution for
multi-project resource and capacity planning in mid-sized industrial
companies: an MCP server built into Rillsoft Project 10, on top of its
scheduling engine – not on top of a task list.
The server runs inside the visible, running Rillsoft Project on Windows and
exposes the open project to an AI assistant or a script over the Model
Context Protocol: project structure and schedule, roles, people, teams,
machines and material, candidate search and staffing assistants, capacity
analysis, baselines and plan/actual comparison, progress and status date, the
resource pool, undo and redo.
This repository contains documentation, client configuration and the npm
launcher (launcher.js, see Install via npx).
The server itself is part of the Rillsoft Project 10 desktop application; the
launcher only forwards to it, and there is no hosted endpoint.
Full documentation: https://rillsoft.ai/en/mcp/
## At a glance
| Endpoint | http://127.0.0.1:3928/mcp |
| Protocol | MCP over Streamable HTTP, revision 2025-06-18 |
| Reach | 127.0.0.1 only – no cloud service, nothing is sent to Rillsoft |
| Requirement | Rillsoft Project 10 on Windows |
| Access protection | optional API key, read-only mode, browser origins rejected |
| Trade-off | an MCP call carries no user identity and reaches exactly one Rillsoft Project instance |
## What it is not
Not a task, to-do or Kanban tool, and not a platform for chat, wiki or team
collaboration.
Not a standalone AI product or chat plug-in, but AI access to Rillsoft
Project as a whole – product, editions and pricing are on
www.rillsoft.com.
Not an AI that schedules on its own: Rillsoft Project calculates dates,
workload and the critical path; the AI reads the result.
Not a cloud service: the server is reachable on 127.0.0.1 only, inside
the open Rillsoft Project. Rillsoft sends nothing; what your AI client sends
is governed by its provider's privacy policy.
Not a team server: there is no remote access.
Not a portfolio editor: portfolios can be opened and evaluated, but not
changed.
## Set up
1. Enable the server. In Rillsoft Project open Settings → MCP server,
set the active switch and leave read-only mode switched on for now.
Alternatively start RillPrj.exe /mcp (port 3928) or
RillPrj.exe /mcp:8123 (another port) for one session.
Check: the status line shows "running on port 3928".
2. Connect your AI client with clients/generic.json:
json
"mcpServers": {
"rillsoft-project": {
"type": "http",
"url": "http://127.0.0.1:3928/mcp"
With an API key set in the settings, use
clients/generic-api-key.json
(Authorization: Bearer ). Ready-made entries per client:
| Client | File | Where it goes | Verified |
| Claude Code | claude-code.mcp.json | .mcp.json in the project folder, or claude mcp add --transport http rillsoft-project http://127.0.0.1:3928/mcp | 12 Oct 2026, 2.1.296 |
| Claude Desktop | claude-desktop.json | claudedesktopconfig.json (Settings → Developer → Edit Config); bridge mcp-remote needs Node.js | 12 Oct 2026, 2.31226 |
| Codex | codex.toml | ~/.codex/config.toml, or codex mcp add rillsoft-project --url http://127.0.0.1:3928/mcp | 12 Oct 2026, CLI 0.162 |