Skip to main content
The Kowalah connector (an MCP server, also called the Kowalah Claude Connector) lets you talk to your Kowalah workspace from inside Claude, ChatGPT, Codex and other AI tools. It connects via OAuth using your Kowalah account and respects the same role-based permissions as the platform. The Kowalah Plugin includes this connector plus 12 skills that know how to use it, so most people should install the plugin. Add the connector on its own if your organisation only allows connectors, or your AI tool doesn’t support plugins. The address is https://mcp.kowalah.com/api/mcp, and Connect Kowalah to your AI tool has the steps for each tool.

​
What you can do

The connector covers nine jobs, across 26 tools:

Get an update

Ask Claude how things are going. Get a portfolio rollup, a focused view of a project, deliverable, Expert Request, or opportunity, and see any discovery invites waiting for your input.

Find an accelerator

Search the global Kowalah library and your organization’s private accelerators for a prompt, GPT, or workflow that already solves your problem.

Raise an opportunity

Capture a new idea, ask, or question for your organisation’s AI lead to triage. Optionally link it to a discovery you’ve been invited to.

Run discoveries and triage

For admins and core team: set up your own discovery, list and triage every opportunity, then promote the shortlist into your own projects or ask Kowalah to take it on.

Check your training

What you’re booked on, your agenda for the day, and your own results, scores, badges, and certificates.

Read your operating model

Your coverage map, a function’s process portfolio, or one process in full: steps, dispositions, quality evals, and the work already aimed at it.

Keep the map true

Edit the processes and steps you own, set targets and quality bars, connect processes, and propose changes or a redesign for the ones you don’t.

Know your AI estate

The systems your work runs on and the AI actually switched on in them: what’s unsanctioned, unowned, acting without approval, or about to renew.

Record the value

Record what AI has delivered on any work, with claimed and verified figures kept apart, and see progress against your vision’s measures.

​
Work your operating model in conversation

The AI Operating Model tools are the biggest thing the connector does beyond reporting status. Ask Claude where AI could help most, drill into a function, talk a single process through step by step, then fix what’s wrong:
If the process is yours, or sits below you in the org tree, the change lands. If it isn’t, Claude offers to propose it to the owner instead, and knows which of those applies before it tries.

Your operating model in Claude

The full walkthrough: reading the map, setting targets and evals, and proposing changes

​
Find the discoveries you’re invited to

When a Discovery is run, by your organisation or by Kowalah, each person on the audience gets a brief asking for their experience of a specific problem. The connector surfaces those briefs to you the moment you ask Claude for an update. Ask Claude:
Claude calls kowalah_get_update and the response includes a pending_discoveries section listing each discovery you’ve been invited to, along with:
  • The discovery’s goal — what Kowalah is researching
  • The business value context — why it matters
  • The close date — when responses are no longer being collected
  • The project the discovery sits against (if any)
You only see the brief that’s been shared with you; you don’t see other people’s responses or the discovery’s internal triage.

​
Submit an opportunity

Anyone in your organization can raise an Opportunity through the connector — admin, core team, or member. You don’t decide upfront whether the work is a deliverable, an Expert Request, or a new project. Your organisation’s AI lead (its admins and core team) triages it first, and Kowalah’s team helps turn the shortlisted ideas into work. Ask Claude:
Claude will:
  1. Check the accelerator library first — there may already be a prompt or workflow that solves it
  2. Walk you through goal, what good looks like, and context Kowalah needs
  3. Call kowalah_create_opportunity with a properly-fleshed-out description
  4. Show you the opportunity number (OPP-XXXX) so you can refer back to it
You can also pick Submit an opportunity from Claude’s prompt menu. It runs the same coaching flow and accepts an optional one-line idea to seed the conversation.

​
Linking an opportunity to a discovery

If Claude has shown you a pending discovery in your update and you want to respond to it, just say so:
Claude passes the discovery’s ID through to kowalah_create_opportunity, which:
  • Stamps the opportunity with the discovery and its project
  • Marks your invite as responded
  • Puts the opportunity in the discovery’s triage queue alongside everyone else’s input
You can also start from the prompt menu with a discovery in mind — Submit an opportunity accepts a discovery ID and pulls the brief into the coaching prompt so the conversation is framed around what Kowalah is actually researching.

​
How it works

  1. The Kowalah connector is added in Claude, either by a workspace admin for everyone or by each person for themselves (see setup in Claude)
  2. Each user signs in with their work email on first use (one-time OAuth). If you don’t have a Kowalah account, one is created, and your email domain places you in your company’s organization
  3. Claude calls Kowalah tools on your behalf, authenticated as you
  4. The server scopes every response to organizations you belong to and the role you hold in each one
The connector is stateless — each request is independently authenticated and scoped. Nothing is cached on Claude’s side.

​
What it can’t do

The connector reads widely, but writes narrowly: it can raise and triage opportunities, run your own discoveries, promote shortlisted ideas into deliverables, maintain your own operating model, keep your systems register and AI estate, and record outcomes. It does not provide:
  • Deleting your records. Processes, steps, evals, units, systems, and AI assets are never deleted. The only removals are a connection between two processes, an arrangement on a system, and an outcome (soft-deleted), and Claude asks before running any of them
  • Editing projects, deliverables, Expert Requests, or contracts once they exist. Promoting an opportunity creates a deliverable, but changes after that happen in the platform. Expert Requests are raised in the platform too
  • Editing the text of an opportunity once it’s been raised (admins and core team can rescore, shortlist, park, reject or merge it)
  • Anyone else’s training record, the cohort roster, or trainer notes
  • File uploads or document management
  • Organization settings or member management
  • Launching a discovery. The connector creates and edits it, then gives you a link to launch it in the platform, which sends the invitations
Operating-model edits are also gated by where you sit in your own org chart: you edit what you own or what’s below you, and propose everything else. See who can change what. For a full interactive experience, use the platform or the Kowalah Agent in Slack, Teams, or Google Chat.
The MCP connector works for anyone who signs in with a work email, though it shows most when your organization has a Kowalah programme behind it.

Set up in Claude

Add the connector to your Claude workspace

Available tools

See what each tool does and how role affects it

Your operating model in Claude

Read the map, set targets and quality bars, propose changes

Security and data access

How the connector is scoped, gated, and verified