KnitKnot
You are reading the agent-optimized layer of this page: the literal markdown we serve to AI crawlers and assistants, shipped in the page source of every visit. KnitKnot finds the claims AI repeats about a company, traces them to their sources, and helps the company correct the record.

# Connect to your AI tools (MCP)

How to reach your KnitKnot workspace from Claude, ChatGPT, or any MCP-capable client, so you can query your benchmark and act on it without leaving your assistant.

KnitKnot exposes a curated view of your authorized workspace through the **Model Context Protocol (MCP)**. This authenticated customer tool is separate from the read-only `/mcp` interface that is rolling out on managed customer domains.

The server lives at `https://mcp.knitknot.ai/mcp/`.

## Connect a client

Open **Settings → API & MCP** in the [console](https://app.knitknot.ai). It provides setup instructions for MCP-capable clients such as Claude and two ways to authenticate:

  • - **Sign in with your account** : the standard OAuth flow. Approve the connection once and your client is linked to your workspace, using the same identity and permissions you have in the console.
  • - **API key** : for scripts, agents, or clients that can't do an interactive sign-in, generate a key and pass it as a bearer token. Treat the key like a password; it carries your workspace access.

Either way, the connection is scoped to your workspace : the assistant sees what you'd see.

## What you can do through it

Customer-facing MCP tools are controlled by an explicit allowlist : deliberately lean, so a model picks the right one without mis-routing. At a high level you can:

  • - **Query your presence** : pull the competitive overview (score, threats, capability gaps) and drill into a single competitor.
  • - **Diagnose gaps** : list issues (what AI gets wrong about you, or where you lose) and open the verbatim evidence behind one.
  • - **Plan content** : list playbooks (the fixes, ranked) and open one for the buyer keywords + ready-to-paste draft brief.
  • - **Act** : update a playbook's status; a change made through your assistant appears in the console.
  • - **Orient an agent** : retrieve workspace context before it starts a task.

MCP and the console use the same authorized workspace data. A supported playbook status change made through an assistant appears in the console.

For the full list of tools an assistant can call : what each does, its parameters, and what it returns : see the [MCP tools reference](/docs/mcp-tools/).

## When to use it

Reach for MCP when your workflow already lives in an AI assistant : for example, to ask which issue needs attention, bring source evidence into research, or move a playbook into progress. For dense review, configuration, benchmark execution, and publishing, the console is still the richer surface.

Next: back to [running benchmarks](/docs/run-and-schedule-benchmarks/), or read how gaps become [issues and playbooks](/docs/issues-and-playbooks/).

Raw mirror of this content: https://knitknot.ai/docs/connect-to-ai-tools.md. Site-wide summary: /llms.txt · full content: /llms-full.txt

Docs navigation
Docs › Guides

Connect to your AI tools (MCP)

How to reach your KnitKnot workspace from Claude, ChatGPT, or any MCP-capable client, so you can query your benchmark and act on it without leaving your assistant.

Updated

KnitKnot exposes a curated view of your authorized workspace through the Model Context Protocol (MCP). This authenticated customer tool is separate from the read-only /mcp interface that is rolling out on managed customer domains.

The server lives at https://mcp.knitknot.ai/mcp/.

Connect a client

Open Settings → API & MCP in the console. It provides setup instructions for MCP-capable clients such as Claude and two ways to authenticate:

  • Sign in with your account : the standard OAuth flow. Approve the connection once and your client is linked to your workspace, using the same identity and permissions you have in the console.
  • API key : for scripts, agents, or clients that can’t do an interactive sign-in, generate a key and pass it as a bearer token. Treat the key like a password; it carries your workspace access.

Either way, the connection is scoped to your workspace : the assistant sees what you’d see.

What you can do through it

Customer-facing MCP tools are controlled by an explicit allowlist : deliberately lean, so a model picks the right one without mis-routing. At a high level you can:

  • Query your presence : pull the competitive overview (score, threats, capability gaps) and drill into a single competitor.
  • Diagnose gaps : list issues (what AI gets wrong about you, or where you lose) and open the verbatim evidence behind one.
  • Plan content : list playbooks (the fixes, ranked) and open one for the buyer keywords + ready-to-paste draft brief.
  • Act : update a playbook’s status; a change made through your assistant appears in the console.
  • Orient an agent : retrieve workspace context before it starts a task.

MCP and the console use the same authorized workspace data. A supported playbook status change made through an assistant appears in the console.

For the full list of tools an assistant can call : what each does, its parameters, and what it returns : see the MCP tools reference.

When to use it

Reach for MCP when your workflow already lives in an AI assistant : for example, to ask which issue needs attention, bring source evidence into research, or move a playbook into progress. For dense review, configuration, benchmark execution, and publishing, the console is still the richer surface.

Next: back to running benchmarks, or read how gaps become issues and playbooks.