Build with AI
Your AI, building in liveknowledge
Connect the AI tool you already use, then ask it what your knowledgebase says. It reads your pages in passages, each with who verified it and when, and suggests edits for a person to accept. Set up takes about a minute.
Your AI tool needs one address:
https://mcp-staging.liveknowledge.app/mcpWorks with Claude, ChatGPT, Claude Code, Codex, Gemini CLI, Cursor, VS Code, Windsurf and any MCP client. Free, with a free account you can make while you connect.
> get_context()
knowledgebaseAcme Handbook
scopesread, suggest
pages2 Always in Context, 3,120 of 8,000 tokens
> search_pages({ query: "when do we deploy" })
pageRelease Runbook
headingWhen We Deploy
verifiedverified, 2026-11-12
stalefalse
passageDeploys go out Monday to Thursday.
> suggest_edit({ page: "Release Runbook",
reason: "The freeze now starts Thursday at 16:00." })
pending1 suggestion, waiting for a person to accept
How it works
Three steps, about a minute
One address, one approval, then you just ask. There is nothing to install or run on your computer.
Step 1
Add liveknowledge
Paste the address into your AI tool, or click once in Cursor and VS Code.
Step 2
Approve
A liveknowledge page opens. Sign in or create a free account, check where access goes, and choose Connect.
Step 3
Ask
Ask for a question your knowledgebase answers, or a page to write. It lands in your account, ready to edit and share.
Connect
Pick your AI tool
Step-by-step for the tools people use most. Choose yours, follow the steps, and approve when liveknowledge asks.
Connect Claude
On the web and the desktop app
Open Settings, then Connectors, in Claude.
Open Claude connectors (opens in a new tab)Choose Add custom connector, name it liveknowledge and paste the address as its URL.
URLhttps://mcp-staging.liveknowledge.app/mcpChoose Connect, then approve on the liveknowledge page that opens.
A connector you add on the web shows up in Claude Desktop too. On a Team or Enterprise plan, an owner may need to add it first.
Connected? Try a promptConnect ChatGPT
Connectors in developer mode
In ChatGPT, open Settings, then Apps & Connectors, then Advanced settings, and switch on Developer mode.
Open ChatGPT settings (opens in a new tab)Choose Create, name it liveknowledge, paste the address as the MCP server URL and pick OAuth.
URLhttps://mcp-staging.liveknowledge.app/mcpApprove on the liveknowledge page that opens, then pick liveknowledge from the tools menu in a new chat.
Developer mode depends on your ChatGPT plan, and OpenAI moves its menus now and then: look for custom connectors in Settings.
Connected? Try a promptConnect Claude Code
One command in your terminal
Add liveknowledge for every project:
Terminalclaude mcp add --transport http --scope user liveknowledge https://mcp-staging.liveknowledge.app/mcpIn a Claude Code session, run /mcp, choose liveknowledge and Authenticate. Approve in the browser.
In Claude Code/mcp
Working in a repository? The liveknowledge CLI is another way in, on the same account.
Connected? Try a promptConnect Codex
OpenAI's coding agent
Add the server:
Terminalcodex mcp add liveknowledge --url https://mcp-staging.liveknowledge.app/mcpSign in, and approve in the browser:
Terminalcodex mcp login liveknowledge
Prefer config files? Add [mcp_servers.liveknowledge] with url = "https://mcp-staging.liveknowledge.app/mcp" to ~/.codex/config.toml.
Connected? Try a promptConnect Gemini CLI
Google's agent in your terminal
Add the server:
Terminalgemini mcp add --transport http liveknowledge https://mcp-staging.liveknowledge.app/mcpIn Gemini CLI, sign in and approve in the browser:
In Gemini CLI/mcp auth liveknowledge
Connect Cursor
One click, or one line of config
Add liveknowledge to Cursor in one click:
Add to CursorOr add it by hand to ~/.cursor/mcp.json:
mcp.json{ "mcpServers": { "liveknowledge": { "url": "https://mcp-staging.liveknowledge.app/mcp" } } }In Cursor’s MCP settings, choose to sign in to liveknowledge, then approve in the browser.
Connect VS Code
GitHub Copilot’s agent mode
Add liveknowledge to VS Code in one click:
Add to VS CodeOr add it by hand to .vscode/mcp.json:
json{ "servers": { "liveknowledge": { "type": "http", "url": "https://mcp-staging.liveknowledge.app/mcp" } } }Start the server when VS Code asks, allow the sign-in, and use it from Copilot Chat in agent mode.
Connect Windsurf
Cascade’s MCP config
In Cascade, open the MCP settings and choose to view the raw config (mcp_config.json).
Add liveknowledge to it:
json{ "mcpServers": { "liveknowledge": { "serverUrl": "https://mcp-staging.liveknowledge.app/mcp" } } }Refresh the servers, then sign in to liveknowledge and approve in the browser.
Connect Any client
Anything that speaks remote MCP
Add a remote MCP server (HTTP, sometimes called Streamable HTTP) with this address:
URLhttps://mcp-staging.liveknowledge.app/mcpIf the tool can sign in to servers (OAuth), it finds liveknowledge’s sign-in by itself: there is no client id or secret to enter.
If it can’t, create an API token in liveknowledge’s Settings, under Account, API Tokens, and send it as a header:
HeaderAuthorization: Bearer lwt_…
Try it
Your first prompts
Connected? Paste one of these into your AI tool, or ask in your own words. Mentioning liveknowledge helps it pick the right tool.
Answer From the Pages
“Using our liveknowledge knowledgebase, answer this: how do we roll back a release? Quote the passage, name its page, and tell me whether that page is verified.”
Start With Context
“Read the Always in Context pages of our liveknowledge knowledgebase before you start, then tell me in five lines what you now know about how this team works.”
Suggest a Fix
“The Release Runbook in liveknowledge still says deploys go out any weekday. Suggest an edit that says Monday to Thursday, with the reason, and leave it for me to accept.”
Report What Is Missing
“Search our liveknowledge knowledgebase for how to rotate the API key. If no page covers it, report the gap so someone can write the page, and do not guess.”
Write It Down
“Write up the decisions from this conversation as a new liveknowledge page under Engineering, with a short summary first and a heading per decision.”
Find What Is Stale
“List the pages of our liveknowledge knowledgebase whose review is due, oldest first, with each page’s owner.”
What it can do
Everything you make in liveknowledge
Your AI gets the same building blocks you have in the editor, and everything it makes is yours to change.
Context First
Hands over the pages your team marked Always in Context before the agent starts, within a fixed budget.
Search in Passages
Finds the passages that match a question, each with its page, its heading and whether it is verified.
Read With Trust
Reads a page as Markdown with who verified it and when, whether its review is due, and what links to it.
Suggest, Not Overwrite
An edit arrives as a suggestion with its reason. Nothing on the page changes until a person accepts it.
Report a Gap
Records a question no page answered, so it shows in AI Knowledge and someone can write the page.
New Pages
Writes a new page from Markdown under the parent you name. It starts unverified, like any new page.
In your control
Safe to connect
Connecting needs a free liveknowledge account, so your agent reads only what you can open. Writing a page yourself never needs one.
You approve every connection
The approval page says exactly which site the access goes to. Nothing connects until you choose Connect.
Read-only, if you prefer
Switch on Read-only access and the AI can find and read your documents, but never change them. Our server enforces it.
Revoke it any time
Each connection is an API token named after the tool, in Settings › API Tokens, and lasts six months. Revoke it and access stops at once.
Your AI does the thinking
liveknowledge runs no AI model of its own here. It checks, lays out and saves what your tool asks for.
If it doesn't connect
The page says the request has expired
A connection request lasts ten minutes. Start the connection again from your AI tool.
It says I have reached my limit of API tokens
You can have ten unexpired tokens. Revoke one in Settings › API Tokens, then connect again.
My tool keeps asking me to sign in
The token it held may have expired (they last six months) or been revoked. Remove the connection in your tool and add it again.
Still stuck? The Connecting AI Tools guide covers every tool in more depth, or get in touch.