Operations, project and finance teams that want agents to work through their existing permissions instead of handing an AI a blanket administrator credential.
PMS MCP
Project & finance management as governed MCP tools
PMS MCP exposes a Project/Task Management System and its Finance module to AI agents as roughly 130 typed, governed MCP tools. Current state is under internal development and code validation.
PMS MCP is in development (Production). It is not yet a finished, generally-available product. What's below describes what it does today, not a roadmap dressed up as a result.
A defined user.
A specific problem.
Direct API integrations scatter tool definitions and identity logic across agents, making calls harder to govern, scope and audit.
Project & finance management as governed MCP tools.
PMS MCP is an MCP (Model Context Protocol) server that puts project management and finance operations in reach of AI agents through a typed, governed tool contract rather than ad-hoc API calls. It speaks both stdio and streamable HTTP, so it works with local agent runtimes and remote clients alike. A controlled "act as a specified user" identity model lets an agent operate under the correct permissions for the person it's acting on behalf of, keeping every call scoped and auditable.
Before → system → after.
Each agent integration calls project or finance APIs directly and carries its own authentication and request logic.
PMS MCP presents roughly 130 typed tools through one governed boundary and acts as the specified user.
Local and remote MCP clients use the same contracts, while each operation remains scoped to the represented user's permissions.
A visible path
through the system.
The steps below describe the current product behaviour from the documented source of truth.
- 01
Connect an MCP client
Use stdio for a local agent runtime or streamable HTTP for a remote client.
- 02
Select the represented user
The controlled act-as-user model applies the permissions of the person the agent is representing.
- 03
Call a typed tool
The client chooses from roughly 130 governed project, task and finance tools with declared input shapes.
- 04
Keep the boundary auditable
Calls pass through the MCP server rather than becoming ad-hoc API logic inside each agent.
Built, not planned.
~130 typed MCP tools
Roughly 130 tools across the Project/Task Management System and its Finance module, each with a typed, governed contract so agents call them predictably.
Two governed domains
Covers both project/task management and finance operations from a single MCP server, rather than separate bolt-on integrations.
stdio and streamable HTTP
Speaks both transports, so it works with local stdio-based agent runtimes and remote streamable-HTTP MCP clients.
Act-as-user identity model
Supports a controlled "act as a specified user" mode, so an agent operates with the permissions of the person it's representing rather than a blanket service account.
Deterministic core.
Provider-agnostic edge.
Deployment, with
the boundaries visible.
Deployment
Run the MCP server over stdio beside a local agent runtime, or expose its streamable HTTP transport to approved remote clients. In both cases, clients use the same typed PMS and Finance tool contracts and the same act-as-user identity boundary.
Limitations
- The documented scope is the connected Project/Task Management System and its Finance module, not a general connector catalogue.
- Act-as-user does not grant blanket administrator access; an operation remains limited by the represented user's permissions.
- The published product facts do not claim support for systems outside the current PMS and Finance domains.
See whether PMS MCP
fits your system.
We'll show what exists today, state the limits plainly and decide whether an early product is the right fit.