Skip to content

MCP — AI assistants

Lynqu ships a hosted Model Context Protocol server, so AI assistants can work your Lynqu account by natural language — “log everyone I met at the expo as leads”, “move Acme to Negotiation and draft a follow-up”.

Endpoint Audience Tools
https://api.lynqu.com/mcp/v2 Organizations (Business/Enterprise seats) Full B2B suite: leads and the people on them, contacts, companies, campaigns, events, team, dashboards, automation, price book, deals and quotes
https://api.lynqu.com/mcp/me Individuals Personal cards, card analytics, profile

Both endpoints speak streamable HTTP with OAuth. In any MCP-capable client (Claude, ChatGPT, Cursor, Windsurf, VS Code, …):

  1. Add a new MCP server / connector and paste the endpoint URL.
  2. Your browser opens a Lynqu consent page — sign in and approve.
  3. Done. No tokens to copy; the OAuth connection can be revoked from Lynqu at any time.

Desktop and CLI clients that register with a loopback redirect (http://127.0.0.1, any port) are supported out of the box.

  • Connecting is open to any signed-in user.
  • Running tools requires Pro+AI, Business, or Enterprise (your plan or your org seat).
  • On the organization server every tool call re-checks your org role and the org’s MCP policy (admins can restrict which roles may use the assistant, or set it read-only). The assistant can never see or do more than you can in the app.
  • All three roles — admin, manager and employee — may use the assistant by default. Account executives are exactly who the toolkit is for: an employee can add contacts and leads, work their own pipeline and follow up on their own deals. What they see stays bounded by the same rules as the app — an employee’s assistant reads only their own book, and team, billing, pipeline and scoring configuration remain manager- or admin-only. Admins who want a narrower surface can drop roles in Integrations → MCP.

Records can be corrected, not just created

Section titled “Records can be corrected, not just created”

Records are captured with whatever is known at the time and completed later — an email found after the fact, a title that turned out to be wrong. Every noun an assistant can create, it can also edit:

  • update-contact, alongside create-contact and get-contact
  • update-lead for a lead’s own details
  • update-lead-note to reword a note in place, keeping its author and its position in the timeline (a voice transcript is never rewritable — its text is a record of what was said)
  • update-campaign for a campaign’s name, description, dates and budget
  • manage-department to rename a department or move it under a different parent (admin only — the org chart is what department-scoped lead visibility is computed from)
  • update-opportunity for a deal’s name, type, probability, expected close date and forecast category

Editing is deliberately narrower than creating in one respect: anything with its own rules keeps its own tool. Reassigning a lead goes through assign-lead, stage and won/lost through update-lead-stage, campaign membership through attach-leads-to-campaign, and a deal’s value through manage-deal-line-item — so that a deal’s total and its line items can never disagree.

An assistant that can create a record it should not have created needs a way to take it back, or the only route out is the app:

  • delete-lead and delete-company — manager or admin. Both are archives, not erasures: a lead can be restored by an admin, and a deleted company’s leads, contacts and deals are kept and simply unlinked.
  • delete-contact — manager or admin, because a contact is shared data that other people’s leads point at.
  • delete-lead-note and delete-lead-document — your own by default, or anyone’s on a lead you can edit. Deleting a document frees the storage it was using.
  • delete-opportunity — renewals and upsells only. A lead’s primary deal follows the lead, so closing it out means moving the lead to a won or lost stage.

For two records of the same thing, merge rather than delete: merge-leads for people and merge-companies for accounts. Merging keeps both halves of the history; deleting one of them throws away whichever half it was holding. merge-companies called with no arguments lists the duplicate pairs the organization currently has, each side with its lead, contact and deal counts so it is clear which record to keep.

create-company will not make a second account for a domain or a name it already has on file — it fails and names the existing one instead. Assistants work through lists quickly, and nothing in the product announces a duplicate account until somebody opens the duplicates review tab.

start-enrichment accepts five subject types. ai_contact and network are yours; lead, org_contact (a person on the organization’s roster) and company (an account) are organization-owned — they draw on the organization quota pool and require active membership.

Enrichment is confirm-before-save: start-enrichment returns a job, get-enrichment-job reports when it is ready, and nothing is written until confirm-enrichment accepts it. Applying is fill-empty — a value someone typed is never overwritten. Enriching a contact fills the person’s own record, including identity links such as LinkedIn; a lead created from that contact afterwards inherits them.

A few actions are reachable in the app but have no MCP tool at all. These are not gaps waiting to be filled — they are actions where an assistant being talked into one is worse than the convenience is worth:

  • Sending, accepting or declining a quote. Assistants can draft and edit a draft; the moment a quote reaches a customer or moves deal value, a person presses the button. A quote that has already been sent cannot be edited or deleted over MCP at all.
  • Validating an event ticket. Marking an attendee present is something a door does, not an agent.
  • Reading someone else’s pay. Compensation is admin-only to read and admin-only to write, one band stricter than performance data — a manager can see a rep’s numbers but never their plan.
  • Reordering a board. Assistants can move a deal to another stage — or to a stage on a different pipeline — but where a card sits within its column is a human gesture on a board they cannot see.

Everything else follows your own permissions exactly.

MCP REST API
Consumer AI assistants Your own code
Auth OAuth consent flow API keys
Plans Pro+AI and up Business and up
Shape ~150 task-level tools Resource endpoints

If you’re building an integration, use the REST API. If you’re wiring an AI assistant, use MCP.