Beginner path2 min readKeep one real product question in mind as you read.
A gentle first run

Get one useful thing working before you configure everything.

Start with one real product question. Specky helps you turn it into something you can review.

Try saying

“Here is what we are hearing from customers. What is the clearest problem to solve first?”

Bring one thing

A product URL, note, or customer quote.

Ask clearly

Say what you are trying to decide.

Review together

Check the evidence before you act.

Specky MCP server

What this is for

The Specky MCP server gives Claude, Cursor, and other Model Context Protocol clients direct access to your workspace. An agent can read confirmed product context before it codes, search customer evidence, and draft work that a person reviews in the PM Inbox.

Connect

The server address is:

https://www.specky.space/api/mcp

It uses streamable HTTP. Add that address to your client and sign in when asked. OAuth clients discover everything from a single request and you pick one workspace on the consent screen.

For a headless client, create an API key in Settings → Developer and send Authorization: Bearer sk_live_your_key.

Try it without an account

The feedback_to_spec tool needs no key. It turns one piece of raw customer feedback into a prioritized insight and a PRD problem statement, limited to 60 calls per hour per IP address. Every other tool needs a Specky workspace. Tools that create work or use AI also need a paid plan or an active trial.

Main tools

JobTool
Load product context before codingget_product_context, get_coding_brief
Search evidencesearch_workspace, read_workspace_item
Save feedbackcapture_signal
Draft work for reviewcreate_document, create_insight, create_decision, create_tickets
Run customer interviewsstart_research_session
Report coding-agent progresslist_agent_runs, update_agent_run, validate_agent_plan, validate_agent_change

Your client lists every tool after it connects. New work starts as a draft, so a person decides what ships.

Related

Ready to apply this?

Start with your own product and keep the first read grounded. You'll create an account before the guided result so it can be saved to your workspace.

Try the guided start