MCP interface: what Claude and other AI clients can do in your workspace

The Project Manager MCP server gives AI clients such as Claude controlled access to your workspace: 114 tools in 9 areas for reading, creating, updating and – with the right permission – deleting data.

Prerequisites
  • You use an MCP-capable client, for example Claude.ai, Claude Desktop, Claude Mobile, Claude Code or Cursor.
  • You are a member of an organization. Which tools you see depends on your role, the access rights of the connection and the enabled modules.

What MCP is and what the server provides

MCP (Model Context Protocol) is an open standard that lets AI clients call tools of an application. The Project Manager server is called project-manager (version 1.0.0), is reachable at the path /api/mcp of your instance and speaks JSON-RPC 2.0 over HTTP (protocol version 2024-11-05). It provides tools only – no resources and no prompts. The AI decides from your request which tool to call with which arguments; the result comes back as formatted JSON.

The nine tool areas

There are 114 tools in total: 55 read tools (scope read), 42 write tools (scope write) and 17 delete tools (scope delete). Every area has its own article that describes each tool with parameters, roles and specifics. The tool catalog lists everything with all requirements.

Organization, members and analytics
4 tools – Organizations and memberships, enabled modules, team members, web analytics (PostHog/GA4).
Tasks, projects and milestones
17 tools – Tasks (including AI effort estimates and similarity search), projects and milestones.
Sprints and weekly planning
10 tools – Sprints with planned/actual hours, AI sprint suggestion, weekly planning and utilization.
Time tracking and calendar
9 tools – Time entries and timer, Google Calendar events, shared free slots across the team.
CRM, Lead-Engine and network
17 tools – Leads and activities, Lead-Engine dashboard and weekly briefing, email drafts, network contacts.
Invoices and accounting
10 tools – Invoices including finalization and correction, receipts, income overview.
Budgeting
13 tools – Project budgets, retainers, burn-rate forecast, expenses and member rates.
Knowledge, notes, meetings and files
19 tools – Knowledgebase with semantic search, notes, meeting notes, file metadata.
Goals, metrics and weekly reviews
15 tools – Goals, KPI metrics and weekly reviews.

What you can do with it

These examples show how requests to the AI translate into tool calls. You phrase things in plain language; the client calls the tools itself.

Tasks
"Create the task 'Review proposal' with high priority in the Relaunch project and assign it to Anna." – list_projects, list_members, create_task.
Track time
"Start a timer for task X" or "Stop my timer." – create_time_entry, stop_timer.
Planning
"Compose a sprint from the backlog and apply the suggestion." – create_sprint, suggest_sprint_plan, add_task_to_sprint.
Sales
"Which leads are overdue? Prepare an email draft for the top three." – list_crm_leads, get_crm_lead, create_crm_email_draft. A human sends it in the CRM.
Billing
"Create an invoice draft from the approved time of the project." – list_time_entries, create_invoice.
Finance
"What is the budget status of all projects, and when will project Y's budget run out?" – list_budget_overview, get_budget_forecast.
Knowledge
"Search the knowledgebase for our onboarding process and summarize it." – search_kb, get_kb_item.
Reporting
"How did pageviews and sign-ups develop over the last 30 days?" – get_analytics.

How a call works

The client talks to the server with plain JSON-RPC requests. These methods are supported:

initialize
Handshake; the server reports name, version, protocol version and that it offers tools.
ping
Reachability check; empty response.
tools/list
Returns the tools available to you with description and parameter schema. For account-wide keys without an organization header all tools are listed with an additional organization parameter.
tools/call
Calls a tool with arguments. The server checks organization, role, scope and module and returns the result as JSON text – or a JSON-RPC error with a message.
notifications/*
Client notifications are accepted and not answered.
  1. The client authenticates with a bearer token (OAuth access token or access key).
  2. The client loads the tools with tools/list; tools you are not allowed to use are missing.
  3. The AI picks a tool and calls it with tools/call.
  4. The server runs the action in your organization and answers with the result.

What is deliberately not possible via MCP

The server does not offer some actions. They remain user actions in the web app.

Sending emails
MCP only creates drafts (create_crm_email_draft). Sending stays a click by a human in the CRM.
Marking an invoice as paid
update_invoice rejects the status paid; record a payment instead.
Changing an issued invoice
After finalization only the status can change. Corrections go through create_invoice_correction; line items cannot be edited later via MCP.
File contents
Files are available as metadata only (list_files, list_file_folders); uploading and downloading go through the web app or WebDAV.
Uploading receipts
Accounting receipts can be listed, updated and deleted via MCP, not uploaded.
Changing calendar events
Google Calendar events can be read and created, not changed or deleted.
Other modules
There are no MCP tools for Chat, Vault, CMS or for organization, role, member and subscription management.
Other people's time entries
Time entries and timers always belong to the holder of the connection.

Responsibility, traceability and AI labeling

All actions run on behalf of the connection holder and with their rights. Invoice actions (create, finalize, status change, correction) and the deletion of accounting receipts are recorded in the audit log with the actor type "mcp". AI-generated results are labeled: email drafts carry the "AI-generated" flag, the weekly briefing returns aiGenerated, and the effort estimate and sprint suggestion are to be treated as AI output. Review such results before you adopt them.

Common pitfalls

  • You see fewer than 114 tools

    The tool list is filtered by role, key permissions and enabled modules. If an expected tool is missing, check these three levels in the article "Roles, permissions and modules".

  • IDs instead of names

    Tools work with IDs. Let the AI call list_* or search_* tools first so it finds the right ID before changing anything.

  • AI results are suggestions

    Effort estimates, sprint suggestions, the weekly briefing and email drafts are AI-generated. Adopt them only after review.

Keep reading

Still have a question or a problem?

Visit support

Last reviewed on 2026-09-26