Documentation contents

MCP reference

Every action a connected AI can take.

A finite tool catalog with account authorization, explicit schemas, and honest read/write behavior.

Start your search
On this page

Access tiers

Read — workspace:read
Inspect account-owned records, source results, skills, files, versions, and saved analysis.
Write — workspace:write
Create or update bounded records after ownership, version, and schema checks.
Approval actions
Financial approvals and offer submission remain separate explicit operations. Tool annotations describe risk but never grant permission.

Companies and listings

Search maintained company or listing sources, inspect a result, and save a server-fetched source snapshot. Source tools never mark a public-record company as for sale.

Cohorts and playbooks

List saved opportunities and apply a chosen skill to explicit immutable inputs. There is no arbitrary batch executor or hidden autonomous session.

Financials and scenarios

Read versions, submit a proposal, approve or reject it, revise scenario terms, and save results against exact basis IDs.

Contacts and brokers

Create one private contact, link it to several opportunities, save an outreach draft, and record work without sending email.

Buyer profile

Read or update the account's geography, industries, economics, experience, and constraints used as search context.

Documents

Begin an original upload, send bounded verified chunks, complete it, and list or retrieve account-owned files.

Pipeline

Create or update an opportunity, change phase explicitly, link a company, or record an offer outcome without rewriting its submitted basis.

Analysis reports

Get a context envelope, submit a versioned skill result with evidence and assumptions, then list or inspect saved reports.

Failure responses

Unknown versions, foreign references, stale bases, missing evidence, failed deterministic checks, and unavailable source selections return explicit outcomes. They do not write partial success or invent a replacement result.