What MCP enables
MCP is the protocol that gives an AI client access to Verstt tools. Describe a task in plain language; the agent can inspect course structure, edit objects and check the result. It can add a question, change a trigger or show a slide snapshot.
Verstt provides a local stdio server: your client starts the installed executable with --mcp. Setup uses a command and arguments, rather than the verstt.com website address. The server, dependencies and runtime are bundled with the app. You do not need to install Node.js, download the source code or run a server on a VPS.
Choose a client that can start local MCP servers, such as Claude Desktop, Claude Code or a local Codex client. For the open course, run the client in the same environment and OS user account as Verstt. Your AI provider has its own access and billing terms; Verstt's MCP support does not include a model subscription.
1. Prepare Verstt
- Install the current Verstt release for your OS and launch it.
- For a first test, open a new blank course or a copy of the editable example.
- Open File → App settings….
- Copy MCP client configuration. It contains the actual command and arguments for your installed copy.
- For the open document, turn on Allow AI agents to edit the open course (MCP). This switch is not needed for direct editing of closed files.
Agent changes to the open course appear immediately. Each MCP edit is an Undo step (Ctrl/Cmd + Z); save the source through the editor as usual.
2. Connect your AI client
Choose one method below. If a verstt server already exists, update its entry instead of adding a duplicate. Preserve other servers and client settings.
Claude Desktop: JSON configuration
Open claude_desktop_config.json. In the Developer settings layout, use Settings → Developer → Edit Config. The official MCP guide lists the config-file locations for each OS. Merge the verstt object into the existing mcpServers object, then fully restart the client.
These examples show the structure. Use the path from your own Verstt settings; replace any YourName or yourname placeholder.
{
"mcpServers": {
"verstt": {
"command": "/Applications/Verstt.app/Contents/MacOS/Verstt",
"args": [
"--mcp"
]
}
}
}{
"mcpServers": {
"verstt": {
"command": "C:\\Users\\YourName\\AppData\\Local\\Programs\\Verstt\\Verstt.exe",
"args": [
"--mcp"
]
}
}
}{
"mcpServers": {
"verstt": {
"command": "/home/yourname/Applications/Verstt-1.18.0-x86_64.AppImage",
"args": [
"--mcp"
]
}
}
}Windows backslashes are escaped as \\ in JSON. On Linux, use a stable absolute path to the executable AppImage: if the generated configuration points into /tmp/.mount_…, replace the command with the AppImage itself. Keep the file at that location. If necessary, allow it to run as a program in its file properties.
After restarting, check the local server status in Claude Desktop settings. Newer layouts list local servers in the extensions area; placement varies by version. See Claude Desktop's help for status and logs.
Claude Code: launch command
In a terminal in your working folder, add the server. This macOS example also works for other platforms after replacing the executable path with your Verstt command. Keep --mcp after the -- separator.
claude mcp add --transport stdio verstt -- "/Applications/Verstt.app/Contents/MacOS/Verstt" --mcp
claude mcp get versttUse /mcp inside Claude Code to inspect servers. If it asks you to trust a server or allow tools, check that it launches your installed copy of Verstt. Official Claude Code instructions.
Codex: CLI or config.toml
If you have Codex CLI installed, add a local server with this command. Replace the path to match your OS and installation.
codex mcp add verstt -- "/Applications/Verstt.app/Contents/MacOS/Verstt" --mcp
codex mcp listAlternatively, add a table to Codex's ~/.codex/config.toml. This file uses TOML; do not paste the JSON mcpServers wrapper into it. Update an existing verstt table instead of creating another one.
[mcp_servers.verstt]
command = "/Applications/Verstt.app/Contents/MacOS/Verstt"
args = ["--mcp"]Restart the local client or begin a new session to reload configuration. Terminal Codex exposes the active MCP servers through /mcp. See the official Codex documentation. For another client, transfer command and args into its local stdio settings.
3. Check the connection
Open a course in Verstt, enable access to it and send the first prompt below. The agent should name the course and outline it without changing anything. Its tool list may include app_status, course_overview, slide_add and other Verstt tools; clients can add their own name prefixes.
Before the first edit, ask the agent to read verstt_guide, the built-in course-model reference. If tools are available but app_status says the app is unreachable, the MCP server itself is already connected: check the running editor, access switch and execution environment.
Two editing modes
| Mode | Course reference | Where changes appear and are saved |
|---|---|---|
| Open course | @app, including an unsaved course. | Changes appear in the editor, with Undo and normal editor saving. |
| Closed file | An absolute .verstt path. | The server writes the file directly; do not edit it in the editor at the same time. |
Creating a file and filling the open editor are different actions. To work in the window, create a blank course in Verstt first and ask the agent to fill @app. The course_create tool creates a separate file and requires a path; it does not accept @app.
For direct file editing, close that document in Verstt first. Before the first edit to an existing file in an MCP session, the server creates a neighboring course.verstt.bak. This is a session backup, not a history of every version. Reopen the file in Verstt after the agent finishes.
The Mac App Store build retains its sandbox restrictions. Prefer opening the document through Verstt and using @app. Direct paths and export destinations must be accessible under the OS permissions. When exporting an unsaved live course, the agent must explicitly specify the output path.
10 example prompts
Copy a prompt, then replace the topic, slide references and paths. The @app prompts assume open-course access is enabled. Count the first slide as 1 in your request; the agent should map it to the actual structure rather than passing your number to tools without checking.
01. Check the connection without editing
Connect to the Verstt MCP server. Read verstt_guide first, then call app_status. Name the open course and show a short outline using @app. Do not change anything yet.02. Create a course in the open editor
Work on the new blank @app course I opened in Verstt. Create “Your first day with the team” for new employees: five 1280×720 slides covering the objective, a decision scenario, an explanation, a single-choice question and results. Use only the attached source text. Ask about missing facts first. Propose the outline, then build it after I approve. Finally validate references and show slide snapshots.03. Create a separate course file
Create a new course at /Users/YOUR_NAME/Documents/Courses/onboarding.verstt. This is an example path; I will replace it with my absolute path before running the task. Do not overwrite an existing file. Topic: volunteer onboarding. Make four English slides: objective, situation, question and results. Use the attached source and leave labeled placeholders for missing images. Save the source, validate it and show snapshots.04. Make the visual style consistent
In @app, restyle slides 2–4 consistently: background #F7F6F2, body text #292732, accent #7357E8 and margins of at least 64 px. Preserve meaning, answers, scores, triggers and branching. List your planned edits first. After editing, show snapshots of these slides and fix overlaps or clipped text.05. Edit a specific passage
In @app, on the slide titled “What to do next”, shorten the explanation to about 60 words while keeping every step and important condition. Do not change other slides, styling, answers or logic. Show the original and revised text, then apply only the version I approve.06. Add a question and passing threshold
In @app, after slide 3 add a single-choice template question: “Where should you ask for help?” Answers: “The support channel”, “Give your password to a colleague”, “Ignore the issue”. The first answer is correct. Set 10 points, one attempt and explanatory correct/incorrect feedback. Add or update a results slide with an 80% passing threshold and include this question. Check both correct and incorrect paths.07. Add a popup explanation
In @app, add a “Details” button to slide 2. Clicking it should show a layer containing “You can ask a follow-up question here” and a “Close” button. Closing must hide only that layer. Preserve the current navigation. Validate the trigger references and show a snapshot with the explanation visible.08. Insert a local image
In @app, add the image from /Users/YOUR_NAME/Documents/CourseAssets/team.png to the slide titled “Meet the team”. I will replace this example path with my actual path. Keep its proportions, preserve important details and avoid the title. Leave other objects unchanged. If the file is inaccessible, report that instead of substituting a random image. Show the result.09. Review a course before publishing
Run course_validate on @app: check slide, layer, object and variable references, duplicate IDs and off-slide objects. Separately inspect snapshots of every slide, feedback layers and button states. List the problems. Fix only clear technical issues and discuss ambiguous choices first. Repeat validation after editing.10. Prepare HTML and SCORM exports
Read the current @app state and validate it. If errors remain, explain them and stop. If there are no errors, export HTML and SCORM 1.2 into a folder I choose. Confirm the absolute output paths before writing, and do not overwrite existing files. List the output files and the results I should check in the target LMS.A useful authoring workflow
- Give context: audience, learning objective, available learner time and the material that should supply the facts.
- Bound the edit: identify slides, the change you want and what to preserve—answers, scores, triggers or style.
- Work in small steps: outline → two or three slides → review → the next section. This makes mistakes easier to spot and undo.
- Ask to see the result: snapshots reveal overlaps, clipping and layer states. Check the interaction itself in Preview.
- Save and test publication: verify scores, completion and resume in your actual LMS.
You can provide local image, audio and video paths. MCP embeds accessible files, but it does not itself generate pictures or synthesize narration: those tasks need separate tools in your AI client.
Troubleshooting
- No Verstt tools appear
- Check stdio transport, the executable path,
--mcp, JSON/TOML syntax and client restart. Do not use verstt.com as an MCP endpoint. Read the client's server startup log. - The server connects but @app is unavailable
- Launch the editor, open a course and enable access in App settings. Use the same user and execution environment. A client in WSL, a container or a remote machine should not be assumed to reach the editor window on the host OS.
- A file is missing or permission is denied
- Use an absolute path and verify it in the OS. Check Windows JSON escaping, Linux AppImage execute permission and MAS sandbox access. Allow the necessary client actions and choose an accessible folder if a destination is blocked.
- A Linux update breaks the launch command
- Check whether the AppImage filename changed. Set a stable path to the new file and reload the client's configuration. A temporary
.mount_…directory is not a permanent launch path. - A conflict or busy error occurs
- Avoid overlapping repeat edits. Ask the agent to reread the course: it may have changed, or you may have switched documents. Do not force a stale copy over the current course.
- The course changed in the wrong way
- Undo in the open editor or restore your own file copy. When asking the community for help, include the OS, app/client versions, editing mode and error text with private information removed.
Capabilities and limits
The agent works with course data and player-rendered snapshots, not a human view of the editor window. A snapshot checks appearance; it does not prove that every click, accessibility requirement or LMS report behaves correctly.
quiz_set_answers is designed for built-in single-choice and multiple-choice templates. Ranking, matching, other templates and imported questions need object-level work and logic checks. Structural validation does not assess the factual accuracy of the learning material.
The MCP connection is local, but model processing depends on your AI client: course text, media and snapshots can be sent to it. Choose source material and provider settings that fit your data requirements. Read about Verstt data handling.
Official references
- Local MCP servers and configuration format
- Claude Desktop: local servers and logs
- Claude Code: MCP commands
- Codex: local stdio servers
Verstt paths, @app, backups and capabilities were checked against the app configuration and bundled help. Third-party client settings may move as those clients are updated.