Revu and MCP

With the Bluebeam Model Context Protocol (MCP) server, you can prompt AI models to run powerful, multi-step workflows in Revu 21.9 or later. These prompts eliminate the need for complex scripting or technical engineering, making advanced automation accessible to Bluebeam Max users.

For information on the tasks you can perform with MCP, see the General FAQs.

To learn about MCPs and how all elements work together, see AI and MCP overview.

This feature is only available with a Bluebeam Max plan.

Connect Revu with AI models to enhance your PDF workflows. To use Revu and the AI model together, you must do the following:

Sign up for the AI model (if required).
Download the AI model desktop application (if available).
Configure Revu and the AI model to allow the MCP connection.

Use the links below to set up Revu and AI models.

Prompts

Prompts are messages that tell the AI model how to perform a specific workflow. Using natural language, you can create prompts that handle tasks such as:

Analyzing documents
Extracting key information
Organizing markups
Creating summaries or reports
Identifying issues or inconsistencies
Rewriting or reformatting content

Use prompts in Revu to:

Automate tedious tasks: Eliminate hours of repetitive work by letting AI handle steps that normally require manual effort.
Improve consistency across teams: Prompts standardize how tasks are performed, reducing errors and variations between individuals.
Lower the barrier to complex workflows: You don't need to be an AI or Revu expert—use natural language to create prompts that simplify advanced functionality into one action.
Boost productivity: More automation → faster output → more time for higher-value work.
Encourage innovation and knowledge sharing: Users can submit and refine prompts together, creating a constantly expanding workflow toolkit.
Build a centralized workflow library: The AI Prompts Library becomes a single source of truth for best practices, reusable workflows, and team knowledge.

Prompts ensure consistency and save time, especially for workflows you repeat often.

Prompt best practices

MCP prompts can unlock complex, automated workflows inside Revu when written clearly and intentionally. Use these best practices to create prompts that are reliable, repeatable, and easy for others to adopt.

Start with a clear goal: Define exactly what the prompt is meant to accomplish. A strong statement helps the AI system interpret your instructions accurately.
- What problem should this prompt solve?
- What output or transformation do I expect?
Be specific: Precise instructions lead to consistent results. Clearly outline:
- Inputs
- Steps
- Output format
- Constraints or rules
Avoid assumptions and spell out every requirement.
Provide context: Include any relevant details, such as:
- File types
- Document structure
- Naming conventions
- Project-specific considerations
Break down complex tasks: If a workflow involves multiple steps, indicate each one. Sequential instructions perform more reliably than single long paragraphs.
Use consistent terminology: Use the same terms throughout the prompt. For example, don't switch between "markups," "annotations," and "notes."
Include the expected output: Tell the AI model:
- What you want
- What it shouldn't do
- How the final result should be structured
This helps both the AI model and anyone reviewing your prompt.
Verify the results: Treat the AI output as a draft that requires human verification. AI can make mistakes, and its work should always be reviewed for accuracy.
Test before sharing: Run the prompt several times to ensure consistent results. If it behaves unpredictably, simplify or clarify the steps.
Create and use Skills: When you've successfully executed a complex prompt that you or your team will want to use again in the future, create a Skill or Agent Flow. See the help documentation for your chosen AI model to learn more.
Document technical requirements: Help others reproduce your workflow by noting:
- Revu version
- MCP configuration
- Any setup required before use
Keep improving: Prompts evolve. Revisit your prompts periodically to refine instructions as tools and workflows change.

Best practices for AnythingLLM

To use the MCP server, invoke Agent mode by typing @agent before your prompt or selecting Start Agent Session at the bottom of the chat window.
Reference MCP tool names directly in the prompt. To learn how to see a list of MCP tools in Anything LLM, see Manage the Bluebeam MCP server in AnythingLLM.
More explicit and deterministic prompts generally produce more reliable results than broad conversational requests.
Complex or repeatable workflows work best as slash commands, saved tools, or reusable prompt templates. This provides more consistent execution than conversational prompting alone.
Avoid large, multi-step flows. AnythingLLM may struggle with large workflows inside a single command. Breaking workflows into smaller steps generally improves consistency and reliability.
Dynamic values, such as current date, workspace variables, or environment variables, should be explicitly referenced in the prompt. Do not assume the model will automatically reference system time or workspace variables correctly unless they are clearly defined.

Tokens

Consider tokens to be the currency of conversation with most AI models. Every word, punctuation mark, image, attachment, etc., is made up of a certain number of tokens. When you send a message or attachment, you "spend" tokens. Longer messages and documents cost more tokens than shorter ones. When the AI model replies to your message, it also "spends" tokens to generate the response. A detailed answer costs more tokens than a simple "Yes" response.

The number of tokens available depends on your chosen AI model and your subscription plan with that AI provider.

Bluebeam can't control the number of tokens used in a conversation, refresh your token budget, or let you know how many tokens you have left.

Some AI models do not use a token system, but may still limit your interactions. See the help documentation for your chosen AI model to learn more.

To learn how to create prompts without spending a lot of tokens, see Top tips for saving tokens.

Top tips for saving tokens

To prompt AI models most efficiently, use these tips for saving tokens:

Start with a clean document: Before prompting, clean up your PDF (run OCR, flatten old markups, or delete unnecessary pages). A cleaner document uses significantly fewer tokens.
Start fresh conversations: AI models reread the entire chat history every time you send a new message. Once you finish a task, start a new chat for your next task. This prevents the old context of the previous task from using extra tokens in your new request.
Be specific and concise: Broad questions often result in longer responses that drain your token budget. Instead of saying "Tell me about this PDF," try "Summarize the top three safety requirements in this PDF in bullet points." Shorter, targeted prompts use fewer input tokens, and specific constraints result in shorter, more efficient output tokens.
Use "selective" OCR: When working with scanned documents, the AI model must process the text version of those images. Perform OCR only on the specific pages you need to analyze rather than the entire document. This reduces the amount of "PDF text" data transmitted through the MCP.
Reduce the size of your file: Referencing a 100-page PDF can cost a lot of tokens. Use the Extract Pages feature in Revu to create a smaller PDF containing only the relevant sections before asking the AI model to analyze it.
Limit the scope of the MCP tools: The AI model doesn't always need to see everything in your file to answer a question. For example, if you only have a question about markups, say "Look only at the Markup List metadata to find the status of the electrical items." This prevents the MCP from extracting and sending unnecessary page text or project file data.
Avoid "chatting" with the AI model: Skip the "Hello, how are you today?" and "Thank you so much!" formalities and go straight to the instruction. Every word, including "Hello" and "Please," counts as a token.
Edit, don't reply: If the AI model gives you a response that is too long or slightly off base, hover over your sent message and click Edit to tweak your prompt instead of sending a new message. This "overwrites" the previous long response and saves your token history from doubling.
Request an outline first: Ask the AI model to "Outline your plan before starting." to prevent the AI model from spending a large amount of tokens on a massive report that was headed in the wrong direction.

For more AI prompt best practices and token-saving prompt templates, see AI prompts overview.

Token-saving prompt templates

In a standard chat interface, use these template instructions to help preserve your token budget.

Quick Summary template
To understand a document or a set of markups quickly, use this prompt:
"In the active PDF in Revu, summarize [insert subject, ex., the structural markups].
- Provide the answer in a bulleted list.
- Be concise: Do not use introductory or concluding sentences.
- Focus only on [insert specific detail, ex., open status items]."

Data Extraction template
To find specific values or dates within a large set of project files, use this prompt:
"Scan the active PDF in Revu for [insert item, ex., all specified concrete areas].
- Respond using a simple table with two columns: [Column A] and [Column B].
- Don't describe the process of how you found the data.
- If the information is not found, respond only with 'Not Found.'"

Issue Spotter template
To find inconsistencies in drawings or markup statuses, use this prompt:
"Check the [insert subject, ex., page labels] for any duplicates or missing sequences.
- List only the errors found.
- Limit your total response to under 100 words.
- If no errors exist, respond with 'No issues detected.'"

Why these work:

AI models typically respond with polite introductions ("Certainly! I have analyzed your document and found..."). Using the "no intro" rule saves about 15-20 tokens per message.
Tables are often more token-efficient than long-form paragraphs because they eliminate filler words.
Telling the AI model what not to do by specifically stating constraints is the most efficient way to save tokens.

The Model Context Protocol (MCP) was built with a "Security-first" philosophy. Connecting Revu to an AI model doesn't give the AI model unrestricted access to your computer or your account. Instead, it creates a narrow, controlled channel that can only access specific data that you request.

For more information, see the Security FAQs.

Disclaimer: Privacy settings and data usage policies for AI models are subject to change by the provider. The information below is accurate as of June 2, 2026. Because these platforms frequently update their interfaces and Terms of Service, we strongly recommend verifying this information directly with your LLM provider.

You're the gatekeeper

The connection between Revu and the AI model is entirely under your control.

The model can't browse your data on its own. It only accesses your data when you give it a specific command that requires that data.
The MCP server runs locally on your machine and only transmits the specific text and metadata required to complete your request.
You control the MCP skills being accessed by the AI model and can approve or deny the AI model to execute commands. Most AI models allow you toggle specific Bluebeam MCP skills on or off and can be configured to ask your permission before running tools or executing commands. See the help documentation for your chosen AI model to learn more.

Data privacy and model training

One of the most common concerns is whether your private data will be used to train future versions of an AI model. This depends largely on your chosen AI model, and you should reference the model's help for more information.

If using Cloud Providers (such as Claude, OpenAI, etc.)

You must manually ensure your privacy settings are configured to prevent training:

Opt out of training: For most AI models, you can disable the use of your data for training and still receive the full benefit of MCP. Regardless of this setting, if you choose to give positive or negative feedback on a response, the entire conversation may be sent to the AI model for training.
Delete conversations: With most AI Models, deleted chats aren't used for future model training, even if you've opted in to training and improving the AI model.
Use incognito chats: Incognito chats (also called Private or Temporary chats) aren't saved to your chat history or to the AI model's memory, and they aren't used for future model training, even if you've opted in to training and improving the AI model. Incognito chats are feature-dependent and may not be available in all AI models.

For more information, see the Security FAQs.

If using Local Providers (such as Ollama, Anything LLM Local Engine)

Platforms like AnythingLLM provide the ability to host and run AI models on your local hardware.

Zero-data leakage: When using a local model, your data never leaves your machine.
No training: Because the model is running on your local machine, there is no external server to receive or train on your project data.

General FAQs

What actions can I perform using AI with Revu?

The MCP server can use the following tools to perform actions in Revu. To learn how to view the tool available in your chosen AI model, see the "Set up" article for that model.

Document information and analysis
get_page_count	Return the total number of pages in the PDF.
get_page_information	Provide page information such as dimensions, rotation, and other properties for each page.
save_as_text	Extract the underlying text content from specific pages or the entire document .
Markup management
add_markup	Create a new markup.
add_markup_capture	Attach an image or media file to an existing markup.
create_markup_thumbnail	Generate a PNG image of a specific markup to preview.
delete_markup	Remove a specific markup.
get_markup_shape	Return the geometric data (points, width, height, rotation) of an existing markup.
list_markups_in_pdf	Return detailed information about markups, including properties such as type, author, comments, colors, position.
set_markup_property	Modify markup properties such as colors, opacity, line width, author, subject, comments, and line styles.
set_markup_shape	Modify the geometry or position of an existing markup.
State and workflow management
export_state_models	Extract the workflow state models, such as review processes, and their states to an external file.
get_markup_state	Retrieve the complete state history for a specific markup.
import_state_models	Load status states and models from an external file.
list_state_models_in_pdf	Show available workflow state models, such as review processes, and their possible states.
set_markup_state	Change the workflow state of markups (for example, from "Open" to "Reviewed" to "Closed").
Content processing
search	Search for text in the document.
stamp	Apply stamp images to pages with control over position, rotation, scale, opacity, and blend modes.
Color management
color_process_analyze	Analyze and identify all colors used in the underlying PDF content.
color_process_modify	Batch change colors throughout the document (useful for updating drawing standards or branding).
Studio Project Search and Studio Sessions
list_studio_projects	List all Studio Projects that the user has access to.
list_studio_sessions	List all Studio Sessions that the user has access to.
open_file	Open and activate the specified PDF file in Revu. If the file is stored in a DMS, it'll be checked out.
studio_project_search	Search the Studio Project specified by the project ID.
Document editing
create_bookmarks	Create bookmarks using page labels or the text contained within a defined region of a PDF.
redact	Redact all areas on specified pages marked with Redact markups.
set_page_labels	Set page labels, including all options for numbering.
Measurements
set_page_scale	Set the scale for your document using a preset or custom scale.
Custom columns
export_custom_columns	Extract the definitions and settings of user-created custom columns in the Markup List to an external file.
import_custom_columns	Load custom columns from an external file into the current document's Markup List.

Security FAQs

When should I use a standard chat vs. an incognito chat?

Incognito chats (also called Private or Temporary chats) are feature-dependent and may not be available in all AI models.

While both chat modes are secure, choosing the right one depends on your goal.

Feature	Standard chat	Incognito chat
Best for	Ongoing projects, documentation, and tasks where you want the AI model to remember context.	One-off questions, sensitive data processing, or starting with a "clean slate."
History	Saved to your sidebar; searchable for future use.	Not saved. Once you close the window, the conversation is deleted forever.
Memory	Uses and updates your personal "Memory" (preferences, styles, and facts).	Isolated. Doesn't use or update your existing memory or preferences.
Model training	Based on your Privacy settings.	Never used for model training, regardless of your settings.
Data retention	Indefinite (until you delete it).	Temporary (deleted from history immediately; purged from backend in 30 days).

When to use Incognito chats:

Processing sensitive data: If you need to analyze a specific log or financial snippet but don't want that data stored in your chat history.
Testing new prompts: When you want to see how the AI model responds without "contamination" from your previous instructions or memories.
Public/shared devices: If you are logged on to a machine where you don't want others to see your recent activity in the sidebar.

Prompts and tokens

How do I create prompts that don't use a lot of tokens?

To prompt AI models most efficiently, use these tips for saving tokens:

Start with a clean document: Before prompting, clean up your PDF (run OCR, flatten old markups, or delete unnecessary pages). A cleaner document uses significantly fewer tokens.
Start fresh conversations: AI models reread the entire chat history every time you send a new message. Once you finish a task, start a new chat for your next task. This prevents the old context of the previous task from using extra tokens in your new request.
Be specific and concise: Broad questions often result in longer responses that drain your token budget. Instead of saying "Tell me about this PDF," try "Summarize the top three safety requirements in this PDF in bullet points." Shorter, targeted prompts use fewer input tokens, and specific constraints result in shorter, more efficient output tokens.
Use "selective" OCR: When working with scanned documents, the AI model must process the text version of those images. Perform OCR only on the specific pages you need to analyze rather than the entire document. This reduces the amount of "PDF text" data transmitted through the MCP.
Reduce the size of your file: Referencing a 100-page PDF can cost a lot of tokens. Use the Extract Pages feature in Revu to create a smaller PDF containing only the relevant sections before asking the AI model to analyze it.
Limit the scope of the MCP tools: The AI model doesn't always need to see everything in your file to answer a question. For example, if you only have a question about markups, say "Look only at the Markup List metadata to find the status of the electrical items." This prevents the MCP from extracting and sending unnecessary page text or project file data.
Avoid "chatting" with the AI model: Skip the "Hello, how are you today?" and "Thank you so much!" formalities and go straight to the instruction. Every word, including "Hello" and "Please," counts as a token.
Edit, don't reply: If the AI model gives you a response that is too long or slightly off base, hover over your sent message and click Edit to tweak your prompt instead of sending a new message. This "overwrites" the previous long response and saves your token history from doubling.
Request an outline first: Ask the AI model to "Outline your plan before starting." to prevent the AI model from spending a large amount of tokens on a massive report that was headed in the wrong direction.

For more AI prompt best practices and token-saving prompt templates, see AI prompts overview.

Token usage quick reference

Action	Token cost
Simple question (new chat)	Very low
Summarizing a 50-page PDF	Moderate
Analyzing a large PDF + long chat history	High
Running multiple MCP tools in one chat	High
Asking a question in a 20-message thread	High
"Count all door markups in the whole project"	Extreme

My tokens ran out. What should I do?

If you hit a usage limit, you must wait for your chosen AI model's refresh period or upgrade your plan.

Bluebeam can't control the number of tokens used in a conversation, refresh your token budget, or let you know how many tokens you have left.

If you've hit your usage limit and are waiting for a refresh, you can still stay productive in Revu. Here's how you can replicate common AI tasks using the tools in Revu:

Instead of asking AI for a summary, use the Markups List in Revu.
- Filter and sort: Select Filter List at the top of the Markups List.
- Create a markup summary: Create a structured report of every annotation in the file, providing the "at-a-glance" view you would usually prompt AI for.
Instead of asking AI "Where is [X] in this PDF?", use the Advanced Search in Revu.
- Text Search: Search the current page, the whole document, or even all open documents at once.
- Visual Search: Search for a symbol (like a specific light fixture or valve) and select Get Rectangle to draw a box around the icon, and Revu will find every instance of it for you.
Instead of asking AI to "Compare these versions," use the document comparison tools in Revu:
- Compare documents: Revu will automatically highlight every difference between two drawings.
- Overlay pages: Stack two or more drawings using different colors for each version.
Instead of asking AI to "Organize these notes," use Spaces in Revu.
- Define Spaces (ex., Room 101, Corridor B) on your drawing, and any markups placed inside that space will automatically be tagged with that location in the Markups List, making it easy to sort by room without AI help.

Troubleshooting

Troubleshooting flowchart

Follow these steps if the AI models isn't responding or can't "see" your Revu documents.

Is the Bluebeam MCP server running?
- No: In Revu, go to Revu > Preferences > Admin > MCP and make sure your chosen AI model is selected.
- Yes: Continue to Step 2.
Are you using the correct AI interface?
- Web browser: MCP won't work in a web browser. You must download and open the AI model's desktop app.
- Desktop app: Continue to Step 3.
Has the tool permission been granted?
- No/not sure: Look at the bottom of your chat window. If you see a "Permissions Request" or small "App" icon with a red dot, click it and select to allow the tool.
- Yes: Continue to Step 4.
Is the PDF visible to the MCP server?
- PDF is flattened/scanned: If the text is not selectable, unflatten the document or run OCR.
- Multiple tabs: Ensure the PDF you want to analyze is the active tab in Revu.
- PDF has security restrictions: Go to Document > Security to see if any restrictions are enabled.
Still not working? Restart the AI model's desktop app:

a. Completely quit the desktop app (either from the tray icon or Task Manager) and restart.

b. Toggle the MCP server off in Revu preferences.

c. Close and re-open Revu.

d. Toggle the MCP server on in Revu Preferences.

e. Relaunch the AI model's desktop app.

Why do my results vary across different chat sessions or documents?

It is common to see slight variations in how AI models respond, even when using similar documents, for the following reasons:

AI-powered intelligence

AI models can analyze PDF content, suggest appropriate actions, and adapt to your workflow.
AI models can troubleshoot issues and adjust its approach when initial attempts don't produce desired results.
Within a single session, the AI model learns from your feedback and doesn't carry knowledge over from previous chat sessions, which can lead to different initial approaches. You may need to re-establish specific constraints or preferences in a new session.

AI model factors

Different AI models (Opus, Sonnet, Haiku) have varying capabilities and may interpret requests differently.
Model updates and improvements occur regularly, which can affect behavior over time.

Document and prompting factors

PDF structure, complexity, and size influence how AI models interprets content.
Changes to a document between sessions may affect prompt outcomes.
The state of the document in Revu (what's selected, current view, existing markups) impacts available actions.
Complex multi-step requests may be handled differently than broken-down tasks.

Environment factors

The following factors can affect how AI models interpret requests:

Desktop version and configuration
Bluebeam Revu version and settings
System resources and performance

If you aren't getting the results you expect, use these diagnostic prompts to help figure out why:

What you want to know	Suggested diagnostic prompts
What the AI model attempted to do and why	"What did you just try to do with that API command?"
What information the AI model needs to complete the task	"What information do you need from this PDF to complete this task?"
What the AI model sees in the document	"Describe what elements you see on page [X] of this PDF."
Why the task failed	"I noticed that didn't work. What prevented that approach from succeeding?"
Why results are inconsistent	"What's preventing you from completing this request consistently over different sessions and PDF documents?"

The MCP connects AI models like Claude to local files, databases, tools, and workflows, enabling them to access key information and perform tasks such as:

Providing read-only data, such as reading a local file or database and providing you with information on those files
Performing actions, such as modifying a markup or changing markup labels

All data access is initiated by your prompt and limited to supported Bluebeam tools.

For more information on what tasks the Bluebeam MCP can perform, see General FAQs.

The Bluebeam MCP server is installed locally and is configured for use by the MCP host installed on your machine, such as the Claude desktop app.

AI and MCP key terms

AI model	A specialized computer program (such as Claude) that can recognize patterns, understand language, and solve problems.
Model Context Protocol (MCP)	An open standard that enables AI models to safely and consistently access data and tools from Bluebeam.
MCP host	The application that you use to interact with the AI model (such as the Claude desktop app).
MCP server	A lightweight program that acts as a bridge between the MCP host and Bluebeam software. It allows the MCP host to see and use specific Bluebeam data and tools.
Resources	Static data (such as text) that the AI model can read.
Tools	Dynamic functions that the AI model can execute. Unlike resources, tools allow the AI model to perform actions, such as "Change markup color" or "Change the subject of markups."
Prompts	The request you type and submit to the MCP host.

How it all fits together

Using Claude for this example, here's how the data flows when you ask a question:

The MCP host (the interface): You submit a prompt in the Claude desktop app.
The AI model (the brain): The Claude AI model receives your request. It realizes it doesn't have your specific data in its memory, so it looks for a tool or resource to help.
The MCP server (the bridge): This small piece of software connects Claude to Revu. It tells Claude exactly what your data looks like (such as a PDF) and what actions it is allowed to take.
Revu (the source): Your actual data—drawings, images, text, markups, etc.—stay securely within Bluebeam. The MCP server fetches only what Claude needs to answer your specific prompt.

Resources

Revu 21

AI and MCP

Prompt AI models to run powerful, multi-step workflows in Revu 21.9 or later.