DealTraq Docs
Tool reference

Tool reference

Every tool and prompt the DealTraq MCP server exposes, what each one reads or writes, and how writes, rate limits, and pagination work.

The DealTraq MCP server (dealtraq) exposes 42 tools and one prompt. Your AI assistant picks tools based on what you ask; you rarely need to name a tool yourself. This page lists every tool. Each group page explains inputs, limits, and example prompts.

All tools

Access shows whether a tool only reads data or changes it:

  • Read: never changes DealTraq data.
  • Writes immediately: the change is saved when the tool runs. There is no approval step.
  • Destructive: writes immediately and can delete, merge, or remove data.
  • Proposal only: creates a proposal that someone approves inside DealTraq. No company data changes during the tool call.

Cost is what the call counts against your rate limit.

ToolGroupAccessCostPurpose
get_dealtraq_identityIdentity and pipelineRead1Show the organization and member you are connected as.
get_organization_thesisIdentity and pipelineRead2Read your fund's active investment thesis.
get_pipeline_structureIdentity and pipelineRead2Explain your pipelines and their ordered stages.
list_companies_in_pipeline_stageIdentity and pipelineRead4List every company currently in a stage.
query_pipeline_metricsIdentity and pipelineRead2Count active pipeline companies, with the definition used.
query_audit_historyIdentity and pipelineRead3See who changed what, and when.
search_companiesCompaniesRead2Find companies by name, website, description, source, or founder.
get_companyCompaniesRead5Read a company's current DealTraq record.
get_company_research_contextCompaniesRead7Gather everything needed to research one company.
list_company_research_submissionsCompaniesRead3List research reports saved with earlier votes.
read_company_research_submissionCompaniesRead5Read one saved research report in full.
list_company_artifactsFiles and meetingsRead4List a company's files and meeting records.
read_meeting_transcriptFiles and meetingsRead5Read a meeting transcript, page by page.
read_company_fileFiles and meetingsRead10Read the text of a deck or document.
read_company_file_pageFiles and meetingsRead10View one PDF page as an image.
read_company_file_rawFiles and meetingsRead15Get the original file, when your assistant supports it.
list_company_action_capabilitiesCompany actionsRead1List every company action and whether it is destructive.
create_companyCompany actionsWrites immediately10Add a company to DealTraq.
add_company_fileCompany actionsWrites immediately20Attach a file (up to 5 MiB) to a company.
execute_non_destructive_company_actionCompany actionsWrites immediately10Run a non-destructive company action, such as adding a note.
execute_typed_company_actionCompany actionsDestructive10Run any company action, including delete, archive, and merge.
execute_company_actionCompany actionsDestructive10Older form of execute_typed_company_action, kept for existing clients.
get_my_inbox_voteCompany actionsRead2Check your personal Inbox vote on a company.
set_my_inbox_voteCompany actionsWrites immediately10Cast or clear your personal Inbox vote.
propose_company_changesCompany actionsProposal only8Propose a profile or stage change for review in DealTraq.
get_company_change_proposalCompany actionsRead2Check whether a proposal is pending, approved, or executed.
list_inbox_review_queueInbox triageRead4List Inbox companies ordered by how much research they need.
get_company_triage_briefInbox triageRead5Read a company's triage status, open questions, and evidence.
submit_company_research_claimsInbox triageWrites immediately8Save cited research findings against triage questions.
list_document_templatesDocument templatesRead2List your house-style document templates.
get_document_generation_kitDocument templatesRead3Get the instructions and layout for drafting with a template.
read_document_template_exampleDocument templatesRead15Open an original example document for a template.
update_document_template_learningDocument templatesWrites immediately10Save confirmed feedback as a new template revision.
list_discovery_programsResearch ScoutRead4List your Research Scout Discovery programs.
get_discovery_viewResearch ScoutRead8Read a program's Research or Market view.
read_discovery_collectionResearch ScoutRead8Page through a program's sources, findings, and other records.
get_discovery_contributionResearch ScoutRead4Read one submitted provisional contribution.
read_discovery_historyResearch ScoutRead6Page through a program's retained history.
read_discovery_payloadResearch ScoutRead8Read a large historical field in chunks.
get_discovery_usageResearch ScoutRead6See a program's usage and estimated cost.
read_discovery_stage_recordsResearch ScoutRead6Page discovery, verification, and evaluation stage records.
submit_discovery_researchResearch ScoutWrites immediately10Save cited research to a program as a provisional contribution.

The server also provides one prompt, review_inbox, which walks your assistant through a full thesis-based Inbox review.

How writes work

Immediate writes and proposals

Most write tools change DealTraq as soon as they run. There is no approval step for create_company, add_company_file, set_my_inbox_vote, submit_company_research_claims, update_document_template_learning, submit_discovery_research, or any of the execute_*_company_action tools. Ask for a write only when you want it to happen.

propose_company_changes is the only proposal tool. It records a profile or stage change as a proposal, returns an approval link, and changes no company data. The change runs only after it is approved inside DealTraq. Use get_company_change_proposal to check its status.

Destructive actions run without confirmation

execute_typed_company_action and execute_company_action can delete notes, founders, and contacts, soft-delete a company, or merge two companies. DealTraq does not ask for confirmation. Your AI client may show its own approval prompt for these tools because they are marked destructive.

Idempotency keys

Every write tool takes an idempotencyKey, a stable value your assistant generates for one intended change (a UUID for update_document_template_learning, a string for the others):

  • Retrying with the same key and the same input returns the original result (replayed: true) instead of writing twice.
  • Reusing a key with different input is rejected.

If a write does not finish within 30 seconds, DealTraq returns a timeout error. The write may still have completed, so retry only with the same key.

Organization and member come from sign-in

During OAuth you choose a DealTraq organization. Every tool uses that organization and your membership. No tool accepts an organization ID, and writes are attributed to you and the AI client you connected. Your access is checked again on every call, so a removed membership or revoked grant fails on the next request.

Rate limits

Each tool call costs between 1 and 20 units (see the Cost column above). In production you can spend 120 units per one-minute window, counted separately for each combination of DealTraq member and connected AI client. The window resets at the start of each minute.

When you exceed the limit, the tool returns a RATE_LIMITED error with retryAfterSeconds and a retryRequest holding the exact call to repeat.

Separately, every tool call times out after 30 seconds.

Pagination and continuation

Large results come back in pages. Instead of asking your assistant to guess the next page, DealTraq returns the exact follow-up call:

  • nextRequest: the next tool call to make, as { tool, arguments }. It is null when there is nothing more to read.
  • coverage.complete: true once every item has been returned. A result is not complete until this is true. Audit history uses pageCoverage.complete instead.
  • Fingerprints: list results include a fingerprint of the set being paged. If the set changes between pages (a company moves stage, a file is added), the result is marked STALE and nextRequest restarts from the beginning so pages never mix old and new data.
  • Read requests: research tools also return ready-made calls for related reads, such as evidenceReadRequests (every readable file and transcript), visualReviewRequest (PDF page images), and rawReadRequest (the original file).

When you ask for a complete answer, such as every company in a stage, your assistant should follow every nextRequest before summarizing.

On this page