Data cleansing
Test interfaces
Prepare pilot
Train key users
Tick off tasks
From status reports, meeting notes or commits: the agent finds the matching tasks and marks them done.
New · MCP server
Your AI agent works directly in PathHub
PathHub has its own MCP server. Once connected, Claude, ChatGPT, Gemini, Copilot and other AI assistants plan, update and review your projects right in the chat, with your permissions and in your language. No API key: add the address, sign in to PathHub, done.
Works with
ClaudeChatGPTGeminiCopilot StudioGrokMistralCursor20+ more
claude · pathhub
You
Create a plan for our ERP rollout: 180 employees, go-live in 9 months.
Agent
Path “ERP rollout” is ready: 7 phases, 36 weeks, 2 new risks.
Example flow, simplified
With the AI you already use
PathHub speaks MCP, the open standard for AI tools. So you work with the assistant you or your company already use, right up to your own self-hosted AI.
Works with, among others,
ClaudeChatGPTGeminiCopilot StudioGrokLe ChatPerplexityClaude CodeCursorGitHub CopilotWindsurfn8n
Including
- ClaudeAnthropic
- ChatGPTOpenAI
- GeminiGoogle
- GrokxAI
- PerplexityPerplexity AI
- Le ChatMistral AI
Including
- Copilot StudioMicrosoft
- Gemini EnterpriseGoogle
- Le Chat TeamMistral AI
- Open WebUIopen source, self-hosted
- LibreChatopen source, self-hosted
Including
- Claude CodeAnthropic
- CursorAnysphere
- GitHub Copilotin VS Code
- WindsurfCognition
- ZedZed Industries
- Clineopen source
- Continueopen source
- KiroAWS
- OpenCodeopen source
- Gemini CLIGoogle
Including
- n8nworkflow automation
- OpenAI APIOpenAI
- Claude APIAnthropic
- Gemini APIGoogle
- Grok APIxAI
- Your own agentsMCP SDKs
Based on vendor information, as of October 2026. Whether custom connectors are available depends on the plan and settings of each provider. Any client that supports MCP with sign-in (OAuth) or an API key works.
How it works
Three steps, no coding.
Add the address
In Claude or ChatGPT, add a custom connector with app.pathhub.ai/mcp. In Claude Code, Cursor or VS Code a short entry is enough.
Sign in to PathHub
Pick a workspace and access level, click “Allow”. No API key, nothing to copy. Already on the Free plan.
Just say what should happen
Your agent picks the right PathHub tools, looks things up, writes changes and reports back.
What your agent gets done in PathHub
Six examples out of 74 tools. Everything you maintain in the app, your agent can maintain too.
Invoice #2026-114 · €3,600
External consulting€3,600 / €12,000
Actuals from cost entries
Book costs
Invoices and orders land on the right budget line. PathHub computes actuals and commitments itself.
Impact →Likelihood →
Capture risks
Add and assess new risks from conversations or documents, with impact and mitigation.
Analysis
Design
Build
Test
Go-live
Generate plans
One sentence becomes a complete plan with phases, tasks, budget, risks and a compliance check.
GDPR records of processingopendone
Involve works councilopendone
NIS2 reporting linesopen
Check compliance
Find open mandatory items, update their status and document evidence across all Paths.
Status report week 42done
Write reports
Run the built-in PathHub agents for status reports, weekly plans or steering committee briefs.
Workflows that used to be manual
Your agent can combine PathHub with everything else it has access to, such as files, code or other MCP servers.
Release tracking (Claude Code)
“Read this week's commits and tick off what's done in the Path Release 3.2. Create tasks for open bugs.”
Take over the minutes
“Here are the steering committee minutes. Add the decisions as next steps and the new risks to the risk register.”
Monday briefing
“What's due in my Paths this week, what's stuck, and which mandatory compliance items are still open?”
Read
Paths, sections, tasks, export, document search, compliance library, templates, quotas
Paths & tasks
Create and update Paths, add tasks, tick off (up to 50 at once), comment, delete
Project content
Phases, risks, stakeholders, milestones, budget items, cost entries, compliance items, next steps
AI & automation
Generate plans in the background, run agents and planning tools, recommendations, ask a Path, webhooks
All 74 tools at a glance
Your agent sees these tools as soon as it is connected. It picks the right one itself, you don't need to know the names.
Read and analyse 14
- list_pathsAll Paths in the workspace with status, duration and target date
- get_pathA Path as overview (progress, budget, compliance) or in full
- get_path_sectionOne section with IDs: phases, tasks, risks, budget …
- get_taskA single task with its phase and comments
- export_pathExport a Path as Markdown or JSON
- list_documentsDocuments connected to the workspace
- search_documentsSemantic search across your documents
- list_compliance_rulesCompliance library by country, industry, size
- list_templatesPre-built project templates
- get_workspace_contextCompany context (industry, size, country, AI hints)
- get_usageUsage and quotas
- list_agentsAvailable AI agents
- list_membersTeam members for assignments
- list_webhooksConfigured webhooks and events
Paths and tasks 9
- create_pathCreate an empty Path, warns about duplicates
- update_pathChange name, status, target date, notes
- add_taskAdd a task to a phase
- update_taskUpdate a task: dates, assignee, phase, done
- update_tasksReschedule up to 50 tasks at once
- shift_phaseShift a phase incl. tasks and following phases
- complete_tasksTick off up to 50 tasks at once
- delete_taskDelete a task
- add_task_commentLeave a comment for the team
Add, update, delete project content 28
| Phases | add_phaseupdate_phasedelete_phase |
| Risks | add_riskupdate_riskdelete_risk |
| Stakeholders | add_stakeholderupdate_stakeholderdelete_stakeholder |
| Milestones | add_milestoneupdate_milestonedelete_milestone |
| Budget items | add_budget_itemupdate_budget_itemdelete_budget_item |
| Compliance items | add_compliance_itemupdate_compliance_itemdelete_compliance_item |
| Next steps | add_next_stepupdate_next_stepdelete_next_step |
| Checklist | add_checklist_itemupdate_checklist_itemdelete_checklist_item |
| Cost entries | add_cost_entryupdate_cost_entrydelete_cost_entry |
| Compliance evidence (link, note) | add_compliance_evidence |
AI and automation 10
- generate_planGenerate a complete plan from a description
- get_plan_jobCheck plan generation status
- run_agentRun status report, weekly plan, risk scan and more
- run_planning_toolCreate RACI, risk matrix, WBS, SWOT and more
- get_recommendationsAI recommendations for next steps
- ask_pathAsk a question about a Path, with document context
- run_project_checkMulti-agent project check: find gaps, propose changes
- apply_project_checkApply selected proposals of the check
- undo_project_checkUndo the applied proposals
- run_compliance_checkRe-run the compliance check, e.g. after changing the country
Sharing, knowledge and notifications 7
- create_share_linkCreate a share link without login, e.g. for the steering committee
- get_share_linkGet the current share link
- revoke_share_linkRevoke the share link
- save_documentSave minutes or notes from the chat as AI context
- get_learningsLessons learned from completed projects
- list_notificationsWhat is new: deadlines, delays, budget warnings
- mark_notifications_readMark notifications as read
Workspace and webhooks 6
- update_workspace_contextMaintain the company context for the AI
- add_workspace_stakeholderAdd a company stakeholder (with approval limit)
- update_workspace_stakeholderUpdate a company stakeholder
- delete_workspace_stakeholderRemove a company stakeholder
- create_webhookRegister an HTTPS endpoint for events
- delete_webhookRemove a webhook
Plus 4 ready-made prompts you can pick directly in Claude & co.: plan_initiative (Plan an initiative), weekly_status (Weekly status), sync_status_update (Apply a status update), compliance_check (Compliance check)
Secure, traceable, under your control
The MCP server is not a back door. It uses exactly the rules of the app.
Your permissions, nothing more
The assistant only sees the workspace you pick at sign-in and has at most your permissions. It never gets your password.
Read, write and AI separated
At sign-in you decide: read only, edit, or additionally AI actions. With read access it cannot change anything.
Built-in limits
Quotas per plan, 120 calls per minute, limits per Path and a lockout after repeated invalid credentials.
Disconnect any time
Disconnect in one click under Developers. An agent only deletes phases that contain tasks with explicit confirmation.
Everything logged
Every call is logged. Changes trigger the same webhooks as in the app.
Honest about data flow
PathHub runs on servers in Germany. What your agent reads is processed by your AI provider (e.g. Anthropic or OpenAI) under their terms.
Setup
The address is all you need: on first use you sign in to PathHub and confirm access. For scripts and automations an API key works too.
Settings → Connectors → “Add custom connector”, enter the name PathHub and this address, then “Connect”. Also applies in Claude Desktop and the app.
https://app.pathhub.ai/mcp
Settings → Apps & Connectors → create a custom connector (developer mode, depending on your plan), enter this address, authentication OAuth.
https://app.pathhub.ai/mcp
Run in your terminal, then open /mcp in Claude Code and sign in.
claude mcp add --transport http pathhub https://app.pathhub.ai/mcp
Add to ~/.cursor/mcp.json or .cursor/mcp.json. Sign-in opens on first use.
{
"mcpServers": {
"pathhub": { "url": "https://app.pathhub.ai/mcp" }
}
}Add to .vscode/mcp.json. Sign-in opens on start.
{
"servers": {
"pathhub": { "type": "http", "url": "https://app.pathhub.ai/mcp" }
}
}For scripts, n8n or clients without a sign-in window: create an API key under Developers and send it as a header.
claude mcp add --transport http pathhub https://app.pathhub.ai/mcp \ --header "Authorization: Bearer phk_live_..."
Detailed guide with all endpoints:API & developersDocumentation
Frequently asked questions
What is MCP?
The Model Context Protocol is an open standard that lets AI assistants use external tools. Claude, ChatGPT, Gemini, Microsoft Copilot Studio, Cursor and many other apps support it. PathHub provides 74 tools through it.
Which AI assistants can I use with PathHub?
Any that support MCP: in the browser and as apps, for example Claude, ChatGPT, Gemini, Grok, Perplexity and Mistral Le Chat; in companies Microsoft Copilot Studio, Gemini Enterprise, Open WebUI or LibreChat; for development Claude Code, Cursor, VS Code with GitHub Copilot and more. Whether custom connectors are available depends on each provider's plan.
Can we use PathHub with our own company AI?
Yes. Platforms such as Microsoft Copilot Studio, Open WebUI or LibreChat add PathHub as an MCP server, even when they run self-hosted models. An admin sets up the connection once, everyone signs in with their own PathHub account and keeps their permissions.
Which plan includes the MCP server?
All of them. On the Free plan you connect your AI assistant with 200 calls per month, Pro includes 2,000, Max 20,000. AI actions such as plan generation additionally count against your workspace's AI quota. API keys and webhooks for your own integrations are available from Pro.
Do I need to know how to code?
No. In Claude or ChatGPT you just add the address and sign in to PathHub. After that you talk to your assistant like to a colleague.
Do I have to connect PathHub in every chat?
No. Once connected, PathHub is available in your chats, in Claude also in the desktop and mobile apps. Depending on the app you switch the connector on or off per chat.
Does this use my AI quota?
Reading and editing don't, they only count as API calls. AI actions such as plan generation or PathHub agents count against your workspace's AI quota. The conversation itself runs on your Claude or ChatGPT account.
Can the agent break something?
It can only do what you allow at sign-in. Read access is enough for analysis. Phases with tasks are only deleted with explicit confirmation, and every connection can be disconnected instantly.
Does it work with ChatGPT?
Yes. ChatGPT connects directly to the MCP server as a custom connector, in developer mode depending on your plan. Alternatively, a GPT action with PathHub's OpenAPI description works too.
Where is my data processed?
PathHub itself runs on servers in Germany. Content your agent retrieves is processed by your AI provider under their terms. Details are in the Trust Center.
More about PathHub
Give your AI agent a project memory.
Sign up for free, add the address in your AI assistant, sign in. Within minutes it works with real plans.