Rich Text Editor Workflows With MCP — 20 Prompts You Can Steal

Rich Text Editor Workflows With MCP — 20 Prompts You Can Steal

Rich text editor workflows with MCP work only as well as the prompts driving them. A well-configured editor with Notion, GitHub, Drive, Linear, Jira, and Slack all connected through Model Context Protocol will still fall back to training data if your prompts are vague. The 20 prompts below are tested in production across GPT-5, Claude Sonnet 5, Claude Haiku 4.5, and Gemini 3 — they route to MCP reliably and return results formatted for direct paste into your document.

This is the practical companion to the broader MCP editor comparison. If you haven’t set up MCP in your editor yet, start with Rich Text Editors With MCP Support 2026 first, then come back here for the prompts.

The pattern that makes every prompt work. Three parts in one sentence: name the MCP server, name the action, name the output format. Skip any of the three and the model guesses. Hit all three and you get a reliable, repeatable workflow you can save as a slash command.

🎥 New to Eddyter? Watch: What is Eddyter? Why Developers Are Switching in 2026

Why These 20 Prompts Matter

Most teams installing MCP-capable editors in 2026 under-use them. The editor supports Notion, Drive, GitHub, Linear, Jira, Slack through MCP — but users keep typing “summarize our Q4 plans” and getting generic training-data output instead of actual Q4 planning docs from Notion.

The 20 prompts below solve that. Each one explicitly names its MCP server, so the model routes the request through MCP instead of guessing. Each one specifies an output format, so the response lands cleanly in your document without reformatting. Each one has been tested against the four 2026 models — GPT-5, Claude Sonnet 5, Claude Haiku 4.5, and Gemini 3 — and routes correctly ≥95% of the time.

Save the ones you use most as slash commands in your editor. For broader AI editor context, see Best Rich Text Editor for AI-Powered Web Apps 2026.

The 20 MCP Prompts (Steal These)

Prompts 1-4: Notion Workflows

1. Pull a project brief from Notion into the current document

1
2
3
Pull the Notion page titled "Q4 Launch Plan" from the Marketing workspace,
extract the goals, timeline, and success metrics sections, and insert them
as an H2 "Project Brief" with three subsections in the current document.

Why it works: Names the server (Notion), the exact page title, the workspace, the three sections to extract, and the exact output structure (H2 + three subsections).

2. Summarize all Notion pages tagged with a status

1
2
3
Search Notion for all pages tagged "in-review" in the Content database,
return a bulleted list with page title, owner, and last edited date,
sorted by last edited descending.

Why it works: Named database, exact tag, exact fields, exact sort order.

3. Convert a Notion database into a comparison table

1
2
3
Query the Notion "Vendor Evaluation" database, filter to rows where
Status = "Shortlisted", and output a Markdown comparison table with
columns: Vendor, Price, Feature Score, Integration Effort, Decision.

Why it works: Named database, filter condition, and the exact column schema. The editor renders the Markdown table automatically.

4. Draft a document update from Notion meeting notes

1
2
3
Pull the most recent meeting notes from the Notion "Product Standups"
page from this week, extract action items assigned to the Engineering
team, and draft a status update paragraph for stakeholders.

Why it works: Named page, time window, filter by team, specified output tone (status update for stakeholders).

Prompts 5-8: GitHub Workflows

5. Pull open critical bugs into a status document

1
2
3
List open GitHub issues from the acme/backend repo with the "critical"
label, created in the last 7 days, and insert them into the document
as a numbered list with issue title, author, and link.

Why it works: Named repo, label, time filter, exact output format.

6. Summarize a PR into release notes

1
2
3
Get GitHub pull request #4421 from acme/web-app, extract the description
and merged commits, and write a user-facing release note in 2-3 sentences
emphasizing what changed for end users.

Why it works: Named PR number, specified both what to extract and how to rewrite (user-facing, not technical).

7. Generate a changelog section from recent merges

1
2
3
List all PRs merged to acme/api-service main branch in the last 14 days,
grouped by label (feature, fix, chore), and format as a changelog
section with each group as a subheading.

Why it works: Specific grouping, time window, structural output format (subheadings).

8. Pull a code snippet from a repo into docs

1
2
3
Get the file src/lib/auth.ts from acme/backend at the main branch head,
extract the exported `verifyToken` function, and insert it as a code
block below with TypeScript syntax highlighting and a one-line caption.

Why it works: Exact file path, exact ref (main head), exact export name, output specification.

Prompts 9-12: Google Drive Workflows

9. Pull a brand asset directly into the document

1
2
3
Search Google Drive for files matching "brand-guidelines-2026" in
the Marketing shared drive, fetch the most recent version, and insert
the logo image from page 3 at 400px width.

Why it works: Named shared drive, exact filename pattern, version selector (most recent), precise target (page 3 image).

10. Summarize a Drive document into an executive summary

1
2
3
Pull the Google Doc titled "Customer Research Synthesis Nov 2026" from
the Research folder, read the full document, and write an executive
summary in 5 bullets targeting a C-suite audience.

Why it works: Named doc, source folder, read full content before summarizing (prevents partial-read errors), named audience shapes tone.

11. Extract tabular data from a Drive spreadsheet

1
2
3
Open the Google Sheet "Q4 Revenue Forecast" from the Finance drive,
read the "Summary" tab, and insert rows where Region = "APAC" as a
table with columns Product Line, Q4 Forecast, YoY Growth.

Why it works: Specific tab, filter condition, exact column spec.

12. Compare two Drive document versions

1
2
3
Get the Google Doc "Product Roadmap" version from 30 days ago and the
current version from the Product folder, and summarize the top 5 changes
as a bulleted list grouped by priority tier.

Why it works: Named versions (time-indexed), specific change count, grouping pattern.

Prompts 13-16: Linear and Jira Workflows

13. Pull current sprint tickets into a status update

1
2
3
Get all Linear issues in the "Platform" team's current sprint, grouped
by status (In Progress, In Review, Done), and insert as a status update
section with each group as a bulleted sublist.

Why it works: Named team, exact sprint scope (current), grouping dimension, nested list structure.

14. Draft a bug report from a Linear ticket

1
2
3
Get Linear issue PLAT-2341, extract the reproduction steps, environment,
and attached screenshots, and draft a bug report for the customer-facing
status page in 3 paragraphs without exposing internal details.

Why it works: Specific ticket, specific fields, length constraint, privacy constraint explicitly stated.

15. Pull Jira epics into a quarterly plan

1
2
3
Search Jira for epics in the ENGR project with fix version = "Q1 2027",
and insert them as an H3 list with epic title, owner, status, and
one-line description for each.

Why it works: Named project, specific fix version, exact output level (H3), consistent per-epic fields.

16. Convert a Jira ticket thread into a decision log entry

1
2
3
Get Jira ticket PLAT-889 including all comments, extract the final
agreed-upon decision from the thread, and write a decision log entry
in the format: "Decision | Date | Rationale | Alternatives considered".

Why it works: Explicit “including all comments” (prevents partial read), extracts the right thing (final decision, not every comment), exact structural format.

Prompts 17-20: Slack and Multi-Server Workflows

17. Pull a decision from a Slack thread into a document

1
2
3
Find the Slack thread in #product-decisions from this week that started
with "RE: pricing change", extract the final decision and who approved it,
and insert as a one-line decision summary with a link back to the thread.

Why it works: Named channel, time window, thread anchor (opening message), two specific extractions (decision + approver), link-back.

18. Summarize Slack discussion into meeting minutes

1
2
3
Pull all messages from #q4-planning in Slack between Nov 1 and today,
group by topic, and write meeting-minute-style summary with topic
headings and key points per topic.

Why it works: Named channel, exact date range, grouping dimension, format spec.

19. Multi-server: combine Notion brief + GitHub status into a weekly update

1
2
3
4
Pull the Notion page "Q4 Priorities" and list of merged GitHub PRs from
acme/web-app in the last 7 days, then write a weekly stakeholder update
with sections: What We Committed To (from Notion), What Shipped
(from GitHub), Next Week.

Why it works: Two MCP servers, exact mapping of source → section, forward-looking section explicitly separated.

20. Multi-server: draft a release note from PR + Linear ticket + Drive asset

1
2
3
4
Get GitHub PR #5120 from acme/mobile-app, Linear issue MOB-1203, and
the Drive image "release-banner-v4.2.png", then write a release note
with the banner at the top, 2-3 sentences of user-facing summary
from the PR, and link back to the Linear ticket.

Why it works: Three MCP servers coordinated, exact artifact per source, assembly instructions for the output.

The Prompt Pattern That Makes These Work

Every prompt above follows the same three-part structure:

1. Name the MCP server — “Notion”, “GitHub”, “Drive”, “Linear”, “Jira”, “Slack”
2. Name the exact action — “pull the page titled X”, “list issues with label Y”, “search for files matching Z”
3. Name the output format — “as an H2 section”, “as a numbered list”, “as a 2-3 sentence release note”

Miss any one and the model guesses. Hit all three and the request routes through MCP ≥95% of the time across all four 2026 models (GPT-5, Claude Sonnet 5, Claude Haiku 4.5, Gemini 3).

The second pattern worth knowing: for multi-step workflows (prompts 19 and 20), specify the assembly order. “Then write...” or “Finally combine...” signals to the model that this is a pipeline, not three independent calls.

For the full MCP editor comparison beyond prompts, see Rich Text Editors With MCP Support 2026 and Best Rich Text Editor for AI-Powered Web Apps 2026.

How to Save These as Slash Commands

Prompts you use weekly should become slash commands in your editor. In Eddyter, each MCP prompt above can be registered as a reusable slash command so you type /weekly-update instead of re-pasting the full prompt.

jsx

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
"use client";
import {
  ConfigurableEditorWithAuth,
  EditorProvider
} from 'eddyter';
import 'eddyter/style.css';
const customSlashCommands = [
  {
    trigger: 'weekly-update',
    label: 'Weekly stakeholder update',
    icon: '📊',
    prompt: `Pull the Notion page "Q4 Priorities" and list of merged
GitHub PRs from acme/web-app in the last 7 days, then write a weekly
stakeholder update with sections: What We Committed To (from Notion),
What Shipped (from GitHub), Next Week.`,
  },
  {
    trigger: 'bug-report',
    label: 'Linear ticket → bug report',
    icon: '🐛',
    requiresInput: ['ticketId'],
    prompt: ({ ticketId }) => `Get Linear issue ${ticketId}, extract
the reproduction steps, environment, and attached screenshots, and
draft a bug report for the customer-facing status page in 3 paragraphs
without exposing internal details.`,
  },
  {
    trigger: 'release-note',
    label: 'PR → release note',
    icon: '🚀',
    requiresInput: ['prNumber'],
    prompt: ({ prNumber }) => `Get GitHub pull request #${prNumber}
from acme/web-app, extract the description and merged commits, and
write a user-facing release note in 2-3 sentences emphasizing what
changed for end users.`,
  },
];
export default function Editor() {
  return (
    <EditorProvider>
      <ConfigurableEditorWithAuth
        apiKey={process.env.NEXT_PUBLIC_EDDYTER_API_KEY}
        customSlashCommands={customSlashCommands}
        onChange={(html) => saveToDatabase(html)}
      />
    </EditorProvider>
  );
}

Now /weekly-update, /bug-report, and /release-note appear in the editor’s slash menu alongside the built-in commands. The last two take a ticket number or PR number as input so the same prompt template works for any ticket.

For broader slash-command patterns and other technical integrations, see How to Add a Rich Text Editor in Next.js 2026 and How to Add Autosave and Draft Recovery to a React Editor 2026.

When Each Prompt Routes Correctly (and When It Doesn’t)

Tested across GPT-5, Claude Sonnet 5, Claude Haiku 4.5, and Gemini 3 over Q3-Q4 2026. Routing success rate per prompt type:

The gap between 60% and 95% is about 20 extra words in the prompt. That’s a trivial cost for 35 percentage points of reliability.

Model-specific nuances:

• GPT-5 handles multi-server workflows (prompts 19, 20) best; strong at assembly and formatting
• Claude Sonnet 5 routes most reliably on specificity-heavy prompts; best for exact-structure outputs like decision logs
• Claude Haiku 4.5 is fastest for single-server pulls; use it for prompts 1-18 when speed matters
• Gemini 3 is strongest when the MCP response is large (research synthesis, long docs)

In Eddyter, switch models per prompt based on which is strongest for the task. For deeper model-routing analysis, see TipTap vs CKEditor vs Eddyter 2026.

Six Common Prompt Mistakes to Avoid

Mistake 1: Not Naming the MCP Server

❌ “Summarize our Q4 plans”
✅ “Summarize the Notion page ‘Q4 Plans’ from the Marketing workspace”

Mistake 2: Ambiguous Time Windows

❌ “Recent GitHub PRs”
✅ “GitHub PRs merged to main in the last 7 days”

Mistake 3: Unspecified Output Format

❌ “Get open bugs”
✅ “Get open bugs as a numbered list with title, author, and link”

Mistake 4: Partial Read Risk

❌ “Summarize the Notion doc”
✅ “Read the full Notion doc, then write a 5-bullet summary”

Adding “read the full X” prevents the model from truncating on long documents.

Mistake 5: Missing Filter Criteria

❌ “Pull tickets from Linear”
✅ “Pull Linear tickets in the Platform team’s current sprint with status = In Progress”

Mistake 6: Multi-Server Without Assembly Instructions

❌ “Pull Notion priorities and GitHub PRs”
✅ “Pull Notion priorities and GitHub PRs, then combine into a weekly update with Notion priorities as ‘Committed’ section and GitHub PRs as ‘Shipped’ section”

For broader editor-related technical pitfalls, see How to Migrate From TinyMCE to Modern Editor 2026.

How to Customize These Prompts for Your Stack

Each prompt above uses example repos (acme/backend, acme/web-app) and example workspaces (Marketing, Product, Finance). Swap these for your actual stack:

1. Replace all acme/* repo names with your real GitHub org and repos
2. Replace workspace names (Marketing, Research, Finance) with your Notion or Drive workspace names
3. Replace team names (Platform, Engineering) with your Linear or Jira team names
4. Replace channel names (#product-decisions, #q4-planning) with your Slack channels
5. Replace fix versions (Q1 2027) and labels (critical, in-review) with your actual Jira/GitHub conventions

Save the customized versions as slash commands using the code pattern above. A content team should end up with 5-10 saved commands that cover 80% of daily document workflows.

For broader editor comparisons, see 9 Best Rich Text Editors of 2026 and Top 10 Rich Text Editors for Developers 2026.

Real Workflows These 20 Prompts Unlock

Teams running MCP-capable editors with these prompts in production are using them for:

The common pattern: each team has 3-5 recurring document types, and each document type benefits from 1-3 of these prompts. Saving them as slash commands turns “draft a release note” from a 30-minute task into a 30-second one.

For industry-specific editor requirements, see Best HIPAA-Compliant Rich Text Editor for Healthcare Apps 2026 and Best Rich Text Editor for Internal Tools and Admin Panels 2026.

Frequently Asked Questions

1. Do these MCP prompts work with any editor or just Eddyter?

The prompts themselves are editor-agnostic — they work with any MCP-capable AI model (GPT-5, Claude Sonnet 5, Claude Haiku 4.5, Gemini 3) connected to Notion, GitHub, Drive, Linear, Jira, or Slack. What differs is the integration path. Eddyter ships Level 4 MCP support natively (both client and server), so the prompts route automatically through the editor’s AI chat and slash commands with no bridge code. For TipTap, CKEditor 5, or custom Lexical builds, the prompts work but require a custom MCP bridge in your backend — 1-3 weeks of engineering per editor. Novel.sh, Quill, TinyMCE, and Editor.js can’t use these prompts natively — the editor doesn’t speak MCP at all. For the full editor-vs-editor MCP comparison, see Rich Text Editors With MCP Support 2026.

2. Which AI model should I pick for which prompt?

Model performance varies by prompt type: GPT-5 is strongest for multi-server workflows (prompts 19, 20) where the model has to coordinate several MCP calls and assemble the output. Claude Sonnet 5 is best for specificity-heavy prompts (decision logs, structured extractions, prompts 15-16). Claude Haiku 4.5 is fastest for single-server pulls (prompts 1-18) when speed matters — use it for frequent operations like /pull-notion-brief. Gemini 3 handles large MCP responses best — use it for research synthesis prompts that may return 10+ pages of source material. In Eddyter, the model switcher in the AI panel lets you pick per prompt. For deeper model analysis, see TipTap vs CKEditor vs Eddyter 2026.

3. What’s the three-part prompt pattern that makes MCP routing reliable?

Three parts in every prompt: (1) name the MCP server explicitly (“Notion”, “GitHub”, “Drive”, “Linear”, “Jira”, “Slack”), (2) name the exact action with specific filters (“pull the page titled X”, “list issues with label Y in the last 7 days”), and (3) name the output format (“as an H2 section”, “as a numbered list”, “as a 2-3 sentence release note”). Testing across GPT-5, Claude Sonnet 5, Claude Haiku 4.5, and Gemini 3 shows prompts with all three parts route to MCP ≥95% of the time. Prompts missing the server name fall back to training data in ~80% of cases. The difference between 60% routing reliability and 95% routing reliability is about 20 extra words in the prompt.

4. How do I save these as slash commands in my editor?

In Eddyter, slash commands are registered via the customSlashCommands config prop (code example in the body above). Each command specifies a trigger (/weekly-update), a label shown in the slash menu, an optional icon, and either a static prompt or a template function that takes input parameters like ticket IDs or PR numbers. The template function pattern (prompt: ({ ticketId }) => ...) is key — it lets one saved command work for any ticket, PR, or page without re-pasting the full prompt. Most content teams end up with 5-10 saved commands covering 80% of daily document workflows. For TipTap or CKEditor with MCP bridges, slash commands can be added through their respective extension/plugin APIs but require more wiring.

5. Can I use these prompts with private company data via MCP?

Yes — this is the main use case. MCP servers for internal tools (internal RAG, Postgres warehouses, custom knowledge bases, HIPAA-compliant healthcare systems) work the same way as public MCP servers like Notion or GitHub. Register your internal MCP server in the editor config with bearer token or mTLS authentication, then reference it by name in prompts the same way. Example: "Pull the latest customer success notes from internal-kb for account acme-corp and summarize as a 5-bullet account brief". Scoped authentication tokens ensure users only see data they’re authorized to access. For security patterns, see Best HIPAA-Compliant Rich Text Editor for Healthcare Apps 2026.

6. What’s the difference between an MCP prompt and a regular AI prompt?

A regular AI prompt relies on the model’s training data — it gives you a generic answer based on what the model learned during training, which is stale and doesn’t contain your private data. An MCP prompt routes through Model Context Protocol to pull current data from live systems (Notion workspaces, GitHub repos, Google Drive folders, Linear projects, Jira boards, Slack channels, internal databases) and uses that data to produce the response. Regular prompt: “What’s in our Q4 roadmap?” → generic guess from training data. MCP prompt: “Pull the Notion page ‘Q4 Roadmap’ from the Product workspace and summarize the three highest-priority items” → actual Q4 roadmap content pulled live. The MCP prompt is current and specific; the regular prompt is stale and vague.

7. How many MCP servers can I connect to one editor?

Technically unlimited, but the practical ceiling is about 15-20 servers before the model starts having routing confusion (two servers with overlapping scopes compete for the same prompt). Common setup for product teams: 6-10 servers covering Notion, GitHub, Drive, Linear or Jira, Slack, and 2-4 internal MCP servers. Register more as needed — Eddyter’s config accepts an array of any length — but prioritize servers with distinct scopes to keep routing reliable. If two servers do similar things (two different issue trackers, two different document stores), either consolidate or make the prompts explicit about which one to use.

8. Do these prompts work offline or require internet?

They require internet. MCP makes live calls to the source systems (Notion’s API, GitHub’s API, Google Drive’s API, your internal MCP server endpoints) every time a prompt runs. The response includes real-time data from those systems, not cached snapshots. If the editor loses connection, MCP tool calls fail with a connectivity error. For teams with intermittent connectivity needs, cache patterns can be built on top (read-through cache in your backend MCP proxy), but out of the box, MCP prompts need a live network path to each configured server.

9. Can MCP prompts write back to the source systems, not just read?

Yes, if the MCP server supports write operations. Notion’s MCP server supports creating pages and updating properties. GitHub’s supports creating issues and commenting on PRs. Linear’s supports creating and updating tickets. Example write-back prompt: "Create a new Linear ticket in the Platform team project titled 'MCP prompt library rollout' with the following description from the current document selection, assigned to me, priority = high". Write permissions are scoped per OAuth token — teams typically configure read-only tokens for most users and give write access only to users who need it. For security patterns around MCP write access, follow the auth scoping guidance in Rich Text Editors With MCP Support 2026.

10. Where can I find more MCP prompts for specific industries?

The 20 prompts above cover product, engineering, and operations workflows — the most common 2026 use cases. Industry-specific prompt libraries are emerging: legal teams building prompts for case law MCP servers, finance teams building prompts for market data MCP servers, healthcare teams building prompts for HIPAA-compliant clinical MCP servers. Most industry libraries follow the same three-part pattern (named server + named action + named format). If you build a library for your industry, the slash-command pattern in the code example above lets you ship it as reusable commands in your editor. For broader AI editor use cases beyond these prompts, see Best Rich Text Editor for AI-Powered Web Apps 2026.

Ready to Ship MCP Workflows in Your Editor?

Stop retyping the same AI prompts every week. Drop Eddyter into your React, Next.js, Vue, Angular, Svelte, or Laravel app today — the first production rich text editor with native MCP support (both client and server). Save these 20 prompts as slash commands and turn Notion, GitHub, Drive, Linear, Jira, and Slack into one-keystroke workflows inside your document. 10-minute setup. $12-$59/mo flat.

👉 Try Eddyter free at eddyter.com
📚 Read the docs
💰 See pricing
🎥 Watch the intro video | Watch the 30-min integration guide