Article Summary
The Project Agent is an AI assistant built into your project workspace. It can search your organisation's library, research topics on the web, pull information from your connected external tools (such as GitHub, Linear, Notion, HubSpot, and Slack), and suggest edits to your RFP responses — all through a simple chat interface. Use it to draft stronger answers faster, improve consistency across your project, and plug into your existing workflows.
Estimated Time
Ongoing use as needed throughout a project
Prerequisites
Project Agent feature enabled (see Feature Flags in your Organisation Settings)
An active project with requirements imported
At least one response selected in the project editor
(Optional) Your personal MCP connections set up in Account Settings → My Connections if you want the agent to pull data from external tools like GitHub, Linear, or Notion
Step-by-Step Instructions
Step 1: Open the Project Agent
Navigate to your project in your dashboard.
Select one or more responses you want to work with.
Open the Project Agent chat panel from within the project workspace.
🗒️ Note: The agent automatically receives context about your project, including the selected responses, their current answers, compliance settings, and your organisation's details. You do not need to paste or re-explain this information.
Step 2: Ask a Question or Give an Instruction
Type a message in the chat input. The agent can help with a wide range of tasks.
Drafting and editing responses:
"Rewrite this response to be more concise — keep it under 150 words and lead with our key differentiator"
"Add a closing line to each selected response that reinforces our willingness to provide further detail"
Searching your knowledge base and the web:
"Search our past RFPs for how we've described our data migration process"
"What have we said about our onboarding timeline and implementation methodology in past projects?"
"Look up the latest NIST Cybersecurity Framework updates and summarise what's changed since version 1.1"
Using attached documents:
"Here's a PDF document of X company's latest annual report. Can you incorporate this into responses 1 to 12 to show how we unlock their 2027 strategy with our solutions?"
"I've attached a battle card of a competitor who is the current incumbent. In responses 15–37, can you please talk about each section and where there is a value-add over their current solution?"
Using connected tools (MCP):
"Check Linear for any open tickets relating to SSO and summarise what's in flight for responses 4–8"
"Find the product roadmap page in Notion and use it to answer the responses about upcoming capabilities"
"Search GitHub for our SOC 2 evidence repo and pull in the latest audit attestation references"
"Look up this prospect's account in HubSpot and tailor the exec summary to their recent engagement history"
The agent will explain what it is doing and suggest helpful next steps as it works.
Step 3: Review Suggested Edits
When the agent suggests changes to a response:
The edit appears in the chat as a suggested edit alongside a brief description of what was changed.
Review the suggested content carefully.
Accept or dismiss the edit from the chat panel.
🗒️ Note: The agent cannot edit responses that are locked (e.g. responses containing multiple response fields or response tables) or approved by a reviewer. If a response cannot be edited, the agent will explain why.
Step 4: Use Attachments for Extra Context (Optional)
If you have additional documents that are relevant to the conversation (e.g. a technical spec, a vendor datasheet), you can attach files directly in the chat. The agent will use them as additional context when answering your questions or suggesting edits.
Step 5: Connect Your External Tools (Optional)
If your Admin has activated MCP connectors (e.g., GitHub, Linear, Notion, HubSpot, Slack), you can connect your personal accounts to let the agent pull real-time data from those services while drafting responses.
Navigate to Account Settings → My Connections
Click Connect on any MCP connector you want to use
Once connected, the agent will automatically use your connected tools when relevant to your request
💡 Tip: You don't need to tell the agent which tool to use — just describe what you need (e.g., "find open SSO tickets") and it will reach for the right connector. But you can name the service explicitly if you want to be sure (e.g., "search Notion for...").
Step 6: Manage Your Conversations
Each conversation is saved as a thread with full message history.
You can return to a previous thread to continue where you left off.
Start a new thread at any time for a separate topic or set of responses.
Delete old threads you no longer need.
💡 Tip: Threads are automatically given a title based on the conversation content, making them easy to find later.
🗒️ Please note: you can only see your own chat threads. Other user's chat threads will not be displayed.
Understanding the Project Agent's Tools
The Project Agent has four tools it uses behind the scenes. You don't need to activate them manually — the agent decides which to use based on your message.
🌐 Web Search: Searches the public internet for up-to-date information beyond your organisation's knowledge base.
→ Used when you ask about external topics, industry standards, or current events.
🗒️ Please note: web search is an admin-controlled feature. If it has not yet been enabled for your organization, please contact your administrator to request access.
🔍 Content Search: Searches your organisation's knowledge base — past responses, company policies, product information, and standard answers.
→ Used when you ask about your organisation's capabilities, past work, or existing content.
✏️ Edit Response: Suggests revised content for one or more selected RFP responses directly inline.
→ Used when you ask the agent to draft, improve, rewrite, or adjust a response to a specific requirement.
📄 Document: Creates a standalone document that opens in its own tab alongside the chat — where you can read, edit, and export it.
→ Used when you ask the agent to produce structured content such as executive summaries, compliance matrices, cover letters, gap analyses, or reports.
Connected tools (MCP) in the Project Agent
🔌 MCP Connectors: Access live data from the external tools your organization has activated — such as GitHub, Linear, Notion, HubSpot, Intercom, Grain, and Slack.
→ Used when you ask about information stored in those services, e.g., open tickets, product docs, CRM records, or meeting notes.
The specific tools inside each connector (e.g., read issues, create issues) are chosen by your Admin — some are read-only, and some can make changes in the connected service.
🗒️ Note: Each user must connect their own account to each MCP connector. If the agent says it can't find your data in an external tool, check that you've connected it under Account Settings → My Connections.
A note on Edit Response vs. Create Document
These two tools serve different purposes:
Edit Response updates an existing RFP response inline — use it when you want to improve an answer to a specific requirement.
Create Document produces a new, standalone document — use it when you need something separate from your RFP responses, such as a summary report or a gap analysis.
If your request is ambiguous, the agent will ask you to clarify which one you need.
📄 Working with documents:
Created documents appear in the Document tab next to the Edit button in the Project Agent panel, so you can switch between the chat and your document without losing context.
From the Document tab you can review the content, make further edits, and export to DOCX.
If you ask the agent to revise a document it already created, it will generate a new version (e.g. "Executive Summary v2", "Gap Analysis v3") so you can compare iterations.
💡 Tips & Best Practices
Getting better results:
Select only the responses relevant to your request — the agent tailors its answers to the selected context
Be specific with instructions (e.g. "Make this more concise and focus on our cloud deployment model" rather than "Make this better")
Ask follow-up questions if the first result isn't quite right — the agent remembers the full conversation
Working with multiple responses:
You can select several responses and ask the agent to edit them all in one go (e.g. "Make all of these responses more concise")
The agent will generate a separate suggested edit for each response
Using search effectively:
If the agent's answer seems generic, try asking it to search your knowledge base first (e.g. "Search our past responses for anything about disaster recovery, then draft an answer")
For time-sensitive or external topics, prompt it to use web search (e.g. "Look up the latest GDPR enforcement actions")
Using MCP connectors effectively:
Name the service when you want to be precise (e.g., "Pull from our Notion product wiki" or "Check GitHub for our latest security policy")
Combine connected tools with content search for stronger results (e.g., "Find open Linear tickets about SSO, then cross-check with what we've said about SSO in past RFPs")
If the agent says it can't access a tool, check that you've connected your personal account under Account Settings → My Connections
Scope and filters:
The agent respects your project's source RFP and tag filters when searching content, so results are scoped to the most relevant material
✋🏼 Common Mistakes to Avoid
Not selecting responses before chatting — the agent needs at least one response selected to provide relevant context and suggest edits
Asking for edits on locked or approved responses — the agent cannot modify responses that contain multiple response fields, response tables, or that have already been approved. Unlock or unapprove them first if edits are needed
Providing too little direction — vague prompts like "fix this" produce less useful results than specific ones like "rewrite this to emphasise our 99.9% uptime SLA"
Ignoring the edit description — each suggested edit includes a brief summary of what was changed. Read it before accepting to understand the intent behind the revision
Forgetting about threads — starting a new thread loses previous conversation context. Stay in the same thread if your questions are related
Assuming MCP connectors work without connecting your account — your Admin activating a connector only makes it available. Each user must still connect their personal account
Need Help?
💬 Live Chat: Available in-app
📧 Email: [email protected] or contact your Success Manager directly for urgent support.
📚 Learning Centre: learn.autorfp.ai/en