Skip to main content
The Quotable MCP server reads your data and can start your Quotable agents. Your AI client does the rest: it reasons over the results, writes files, and, if you’ve connected them, uses other tools such as Slack, Notion or Google Drive. These recipes show how the pieces fit.
The only thing Quotable’s MCP server can change is starting an agent run, with run_agent. It can’t send messages, post to Slack or write documents, and an agent with steps that act outside Quotable runs only from the app. Any step that writes somewhere else uses a connector or tool in your AI client, set up and authorized by you, under that tool’s own permissions and terms. Quotable doesn’t provide, operate or see those connectors.

Weekly visibility brief to Slack or Notion

Turn a week of tracking into a short update for your team. You need: a key with read:projects and read:visibility, and a Slack or Notion connector in your AI client (for example, a connector added in Claude).
1

Pull the numbers

2

Add the sources

3

Write and send the brief

Your client asks for permission before it uses the Slack connector. To publish to Notion instead, ask it to create a page in your team’s wiki.
Review the brief before you approve the post. Your assistant writes it from the numbers, and you’re the one sending it.

Outreach list of cited sources

Find the publications and communities AI engines rely on, and turn them into a list your PR or content team can work through. You need: a key with read:projects and read:visibility. Optionally a spreadsheet connector, or a client that can write files, such as Claude Code or Cursor.
1

Collect cited pages

2

Classify and prioritize

3

Save it

In Claude Code or Cursor, the client writes the file in your project. With a spreadsheet connector, ask it to create a sheet instead.

Content briefs in your repository

If your site lives in a repository, let your coding assistant draft briefs next to your content. You need: a key with read:projects and read:visibility, in Claude Code, Cursor or VS Code.
1

Find the gaps

2

Ground each brief in what AI cites

3

Write the briefs

Citations are reported per brand, not per prompt, so your assistant matches pages to prompts by topic. Treat its matches as a starting point.

Scheduled report from the command line

Claude Code can run a prompt without an interactive session, which lets you schedule a report with cron or your CI system. You need: Quotable added to Claude Code (see Claude Code) on the machine that runs the job, with a key that has read:projects and read:visibility.
--allowedTools pre-approves the three Quotable tools this report needs, so the run doesn’t stop to ask for permission. Name the tools rather than allowing all of mcp__quotable, which would also pre-approve run_agent. Use a dedicated key for automation, and revoke it when you retire the job.

Run an agent from your client

Start one of your Quotable agents from a conversation and read its result when it finishes. You need: a connection with read:projects, read:agents and run:agents, or an API key. Agents must be turned on for your workspace.
1

Pick the agent

2

Start the run

Your client asks before it calls run_agent. The run uses your workspace’s credits. Agents with steps that act outside Quotable, such as posting to Slack, can’t be started here; start them in the app.
3

Read the result

A run that waits for someone to approve a step shows as paused until it’s approved in the app.

Agency rollup across brands

If you manage several brands in one workspace, compare them in one pass. You need: a key with read:projects and read:visibility, limited to the brands you report on.
Each brand is at least one tool call, and every call counts toward the key’s limit of 300 requests per minute. A rollup across dozens of brands fits comfortably.

Next steps

Example prompts

Single prompts to start from.

Troubleshooting

Errors and what they mean.

Tools reference

What each tool returns.
Last modified on November 18, 2026