While the exact steps to add an MCP connection will vary depending on your AI tool of choice, they will all ask for our MCP endpoint (https://mcp.dealroom.io/mcp) during the setup process.

After providing that URL during the setup within your AI tool, a browser tab will open prompting you to log in to your DealRoom environment. Once you've successfully logged in and authenticated the connection, your AI will now have access to your DealRoom environment based on your level of DealRoom permissions.

The security model is the same as DealRoom’s public API. Data is transmitted to the AI the customer is using (Claude, ChatGPT, etc.) over HTTPS under their existing DealRoom permissions in the web interface.

DealRoom AI is purpose-built for M&A and keeps all data encrypted within DealRoom. MCP transmits data to external AI tools — similar to using DealRoom’s public API. Both have audit trails (write-only, not read) and inherit the user’s existing DealRoom permissions.

No files are downloaded to the user’s machine; the AI accesses content directly through DealRoom’s API. Document content is transmitted to the AI provider (Claude, ChatGPT, etc.) as part of the query, similar to an API read, and respects DealRoom’s content-protection controls — redaction, watermarking, and limited access alike. For files where the user has restricted access, only metadata is transmitted.

Enterprise/business Claude accounts have training data usage disabled by default at the admin level. Personal Claude accounts may have model training enabled by default, so users must manually toggle it off in account settings. We recommend customers use their enterprise AI accounts, not personal ones, when connecting to DealRoom.

Yes — this is one of the clearest value props for security-constrained customers. MCP eliminates the download-upload workflow entirely. Documents are read in context within DealRoom’s permission structure. No file sits on someone’s laptop. For customers who’ve been told "don’t do that" but are doing it anyway, this is the compliant path.

Data Deletion: MCP never deletes anything; deletion is excluded by design (no rooms, deals, documents, tasks, etc.).

Permissions: All actions respect the acting user’s existing DealRoom permissions. The MCP cannot read or change anything that the user themselves couldn’t.

Write back means pushing structured data back into DealRoom — findings, diligence requests, deal records, Q&A responses. It does NOT mean editing documents (like editing a Word file in the viewer). The AI reads deal context, produces structured output, and writes it back into the DealRoom M&A Operating System where the team already works.

By design, the current beta does not support destructive actions. Scope is Create, Read, Update — not Delete. No deleting documents, deals, requests, or audit records. Bulk export and admin-level operations (user management, billing) are also out of scope.

Yes. Pull the list into AI via the MCP and ask it to analyze for redundancy, propose consolidation, or regroup items by category. The AI makes suggestions — you approve before anything changes in DealRoom.

Yes, with good prompting. Give AI context, as in: upload your M&A strategy into the conversation. Without a clear strategy prompt, the AI will generate an answer, but it may not reflect your specific priorities. Good prompt: "Based on [our M&A strategy document], review our active pipeline in DealRoom and recommend the top 10 targets to prioritize and why."

Yes, AI can use the MCP to analyze your full pipeline and produce formatted reports, executive summaries, and full slide decks. Output is generic by default, but teams can add their own templates to get something nearly presentation-ready.

Yes. Ask your AI to analyze documents in a data room and push resulting findings directly into DealRoom, linked to the relevant deal with source citations.

No. MCP reports are ad hoc — they reflect data at the time of the query. For persistent, live dashboards, BI reporting is still the right tool.

Yes — this is one of the most powerful aspects of the MCP architecture. If a customer has other tools with MCP support, they can call all of them in a single AI conversation. Example: "Search DealRoom for our active targets, enrich with PitchBook data, and find 10 similar companies via Apollo." All in one prompt, with relevant data flowing back into DealRoom.

For one-off questions, yes. Users can query deal and room data through the MCP without a BI seat. For live dashboards your team monitors regularly, BI reporting remains the right tool. The MCP complements DealRoom’s BI reporting.

The MCP has access to all data already in DealRoom — including everything the Outlook integration has synced in (email threads, contacts, attachments, deal updates). Richer Outlook integration = richer MCP outputs. More context in DealRoom = better AI analysis via MCP.

Yes. Any data Outlook has pulled into DealRoom is available to the MCP — email threads, contacts, attachments. All accessible to the AI agent without additional action.

Tools are the individual actions the MCP server makes available to an AI harness (Claude, ChatGPT, etc.). Each of the below DealRoom MCP tools maps to a specific operation in DealRoom, such as listing deals in your pipeline, retrieving documents from a room, or creating a task on a request list. When you prompt your AI, the assistant selects the tools it needs, calls them against your DealRoom account, and works from the results it gets back.

DealRoom continues to refine existing tools and add new ones, guided by product strategy and customer feedback.

list_deals

The primary deal query tool. Filters deals by any field: phase, name, custom field value, last activity date. Uses the same filter syntax as HubSpot's search: bare value = equals, dict with op/value = range/in/contains. Supports order_by (prefix - for descending), field projection via fields, and a limit up to 5000. No offset pagination: if results are truncated, narrow the filters. Custom fields are addressed by their fld_... ID, not their label. Call list_fields first to discover those IDs.

get_deal

Fetch the full detail record for a single deal by ID. Returns contacts, metadata, attachments, phase, custom fields, and the associated room ID if one exists.

save_deal

Create or update a deal. Pass deal_id to update; omit to create. Requires name on create. phase_id comes from list_phases; custom field values go in custom_data as {field_id: value}. Recommends checking list_deals first to avoid duplicates.

list_deal_notes

Return all notes attached to a deal.

save_deal_note

Create or update a Markdown note on a deal. Pass note_id to update an existing note; omit to create. Supports headings, lists, bold/italic/strikethrough, code blocks, and links. Images, tables, and HTML blocks are not supported.

list_phases

Returns all pipeline phase definitions (id, name, color). Use phase IDs when creating or updating deals.

list_fields

Returns all custom field definitions across the org: for both Deal and DataRoom object types. Each field includes its fld_... ID, label, type, and valid options/validation rules. This is the lookup table you need before filtering deals by custom field or writing custom data.

list_rooms

Returns all deal rooms in the organization: id, deal_id, name, and URL for each. This is the starting point for any room-scoped operation.

save_room

Creates a new deal room for a deal. Each deal can have at most one room. Takes the deal_id; verify the deal doesn't already have a room before calling.

get_room_permissions Returns the current user's access level in a room: not_a_member, member, or admin. Always returns a result: never 404s. Call this first before any room-scoped operation to confirm access.

list_room_folders

Returns the full folder hierarchy of a room's VDR. No server-side filtering: all folders come back in one call. Save to a file for large rooms before processing.

save_room_folder Create or update a folder. Requires name and parent_id on create (use list_room_folders to find the parent). Pass folder_id to rename an existing folder.

list_room_documents

Returns all documents in a room. No server-side filtering. For content search, use search_documents instead.

get_document_content Returns the full extracted text of a document. Can be large: save to file rather than loading into context directly. Use search_documents to find relevant documents first.

search_documents

Full-text search across documents in a room. Returns ranked results with snippets showing matched passages. Requires at least 2 characters. The right first step before get_document_content.

create_document_from_org_file

Copies a file attached to a deal (teaser, NDA, CIM: from get_deal → attachments[].file.id) into a room folder as a full document. Bridges sourcing-stage files into a due-diligence room.

list_room_categories

Returns all task categories (request lists) in a room. Categories are the grouping layer above tasks: e.g., "Legal", "Finance", "HR". No server-side filtering.

save_room_category

Create or update a task category. Pass parent_id to nest it under another category; omit for a top-level category. Pass category_id to rename.

list_room_tasks

The primary task query tool. Filters by key, name, category, status, priority, or description content. Same filter/sort/project syntax as list_deals. Returns related folders and documents as ID lists (informational only: not filterable). Up to 5000 results.

save_room_task

Create or update a task. Requires name, category_id, and status_id on create (list_room_categories and list_statuses supply those IDs). Supports priority (low/medium/high) and a Markdown description_md. Pass task_id to update.

list_room_task_comments

Returns all comments on a task in one call. No pagination: task discussions are assumed small enough to fit in context.

save_room_task_comment Post a new comment on a task, or edit an existing one, in Markdown. Pass comment_id to overwrite; omit to create. Visibility follows the room's default for your group: not configurable here.

update_task_documents Link or unlink documents to a task via add/remove arrays of document IDs. Does not create or delete documents.

update_task_folders Link or unlink folders to a task via add/remove arrays. Does not create or delete folders.

list_statuses Returns all task status definitions (id, name, color, type). Use these IDs when creating tasks. Not for findings: use list_finding_statuses for those.

list_room_findings The primary findings query tool. Filters by title, severity, likelihood, type, status, or description content. Severity and likelihood are integers (1/2/3) not labels: filter on the number, map back to Low/Moderate/High and Unlikely/Likely/Very Likely when reporting. Related tasks, categories, folders, and documents come back as ID lists (informational only). Up to 5000 results.

save_room_finding

Create or update a finding. Requires title, type_id, and status_id on create. Supports severity (1–3), likelihood (1–3), and three Markdown body fields: description_md, mitigation_plan_md, and actual_mitigation_md. Pass finding_id to update.

update_finding_categories

Link or unlink task categories to a finding via add/remove. Does not create or delete categories.

update_finding_documents Link or unlink documents to a finding via add/remove. Does not create or delete documents.

update_finding_folders

Link or unlink folders to a finding via add/remove. Does not create or delete folders.

update_finding_tasks

Link or unlink tasks to a finding via add/remove. Lets you connect the diligence request that surfaced a risk to the finding record itself.

list_finding_statuses

Returns all finding status definitions (id, name, color, type). Use these when creating findings: separate from task statuses.

list_finding_types

Returns available finding types (id, name, color): e.g., Risk, Issue, Observation. Required when creating a finding.

DealRoom AI is purpose-built M&A intelligence built into DealRoom. It runs automatically in the background — scanning documents, organizing data rooms, updating deal cards, generating summaries. No prompting required. Designed for structural, repetitive work.

DealRoom MCP is a connector that lets your team access and manage DealRoom data through the AI tools they already have (Claude, ChatGPT, Copilot). Use cases include deep analysis, cross-system workflows, and ad hoc queries using the AI tools you use daily.

They’re complementary, not competing. DealRoom AI keeps your platform current automatically. DealRoom MCP lets you use powerful external AI on that same data for deeper work. DealRoom AI feeds DealRoom with clean, structured deal data → DealRoom MCP lets Claude/ChatGPT use that data for deeper analysis without duplication or re-entry. 1 + 1 = 3.