Kairon MCP setup

Resolve a human-readable Sales Navigator filter value (location, industry, function) to a LinkedIn id, echoing the canonical label so a wrong match is visible. Cached globally; pass the returned urn to search_sales_navigator, or let it resolve names for you.

Read one ICP and its latest (active) version by icpId.

List this org's ICPs, each with its latest (active) version's criteria and conditions.

Takes no parameters.

Fetch a LinkedIn person OR company through the caller's connected account — whichever the target addresses. Address it with target — { type, value } where type is url, public_identifier, or provider_id (pass whichever you have). The kind is detected from the target: a /in/ URL or a bare slug is a person, a /company/ URL is a company. Pass kind: 'company' for a company addressed by universal name or numeric id. Returns { kind, profile | company }; complete data, metered + cached (refresh: true forces a live read).

Add members to a list ({ asset, id, items }, ≤500 items). Each item is { memberId } (a lead or company already in the org), { url } (a pasted LinkedIn /in/ or /company/ URL — INSTANT and FREE, zero provider calls: creates or converges on a bare member that can be resolved later), OR { snapshot } (a search hit — pass the hit as-is, providerId required — which creates/dedups a born-resolved member so the saved list shows the person or company, not a bare id). Per-item results: added | already_member | invalid | not_found. Re-adding a current member converges (already_member). This is how you assemble an audience after a search: pass the result hits straight in.

Create a named list — the campaign targeting unit. asset: 'leads' for a list of people, asset: 'companies' for a list of companies (company-first targeting: build the company set, then find people inside it — pass the finished list's id to search_sales_navigator as companyLists to do that second step). Names are unique per org and per asset, case-insensitive; a taken name fails with the existing list id (no silent merge). Returns the new list with a member count of 0. Build membership with list_add.

Soft-delete a list by asset + id: it disappears from list_list with its memberships, the member leads/companies are untouched (they stay in the org pool, in their other lists, with their qualification state intact), and the name is immediately free for reuse.

One list by asset + id: its name, live member count, and created date. Out-of-org or deleted ids are not found. To read who is IN it, use list_members.

Read this org's lists of one asset, keyset-paginated, newest first, each with its live member count. q matches the list name (case-insensitive). Page with the returned cursor. Deleted lists never appear.

One keyset page of a list's members ({ asset, id, q?, cursor?, limit? }), newest-added first. Each row is { id, title, subtitle, location, linkedinUrl } — the id removes it with list_remove, the URL reads it with linkedin_fetch. q filters case-insensitively over the member's display fields (name, headline, company, location). Page with the returned cursor.

Remove members from a list ({ asset, id, memberIds }). Deletes only the membership — the leads/companies stay in the org pool and in their other lists, and can be re-added later. Returns how many memberships were removed.

Rename a list ({ asset, id, name }). Same uniqueness rule as create: a name already taken by another live list of that asset fails with that list id.

A page of a profile's recent posts (transient — not persisted). Address the person with target — { type, value } (url, public_identifier, or provider_id). Use cursor to page; one page is one metered action.

Classic LinkedIn people search — works WITHOUT a Sales Navigator seat. Pass keywords and/or structured filters: id-resolved filters ({type,query} over location/industry/company/past_company/school/service — pass human-readable NAMES, resolved to ids server-side — a name LinkedIn does not recognize REFUSES the search with CHANNEL_ACTION_FILTER_UNRESOLVED naming it, rather than running without that filter and quietly returning a wider audience; re-issue with a name LinkedIn uses, or drop the filter if you meant the broader search), networkDistance (1|2|3), profileLanguage (ISO codes), openTo, connectionsOf/followersOf (member ids from people.suggest), advancedKeywords (first/last/title/company/school). Or paste a search url to run it verbatim. Transient hits + a cursor; one page is one metered action. To persist a hit, call profile_fetch on it.

Search LinkedIn Sales Navigator over people OR companies. This is the expert path: prefer it over classic keyword search whenever a structured filter fits the intent — filters are precise where keywords are noisy.

Find people by what they DID, not by what their profile says — the behavioural counterpart to search_people / search_sales_navigator. Pick a signal: • engaged_with_profile — who commented on or reacted to a profile's recent posts. config: { target: { kind: "person"|"company", url } }. • engaged_with_keyword — who engaged with posts matching a keyword. config: { keyword, datePosted?, firstDegreeOnly? }. • posted_with_keyword — who WROTE posts matching a keyword. Same config. • recently_connected — the seat's own newest connections. No config. • viewed_my_profile — who viewed the seat's profile. No config; needs a paid LinkedIn seat (Premium, Sales Navigator, or Recruiter). keyword is a full boolean query (OR/AND/NOT, quoted phrases, parentheses). Results are RAW: real people who really did this, with no ICP screening and no qualification — screen them yourself, or add them to a list with list_add. Page with the returned opaque cursor; each page is metered. For "changed jobs" or "posted recently", use search_sales_navigator (changedJobs / postedOnLinkedin) instead — those are Sales Navigator spotlights, not signals.

THIS STARTS SENDING MESSAGES TO REAL PEOPLE ON LINKEDIN, and a message already sent cannot be un-sent. Activates a draft or paused campaign — { id, startNow? }; startNow (default true) enrolls today's first batch immediately instead of waiting for the campaign's next scheduled send window. Rejected with the machine code CAMPAIGN_ACTIVATION_BLOCKED (409, params.unmet listing every unmet precondition as structured data an agent can branch on) unless ALL FOUR hold: no_published_version — a published graph exists (campaign_set_graph then campaign_publish first); no_bound_list — at least one lead list is bound (campaign_bind_list); no_send_window — the campaign has a send window configured; sender_not_connected — its sending seat is still connected. Calling activate on an ALREADY-active campaign is a no-op that returns the current campaign unchanged — safe to retry after a lost response. Sending itself is throttled by the org's own daily caps and send window; this tool does not need to reason about them. 404s (COMMON_NOT_FOUND) if the campaign is not in this org.

Create a draft campaign: a name, the sending seat (channelAccountId — a connected LinkedIn account in this org), and optionally the ICP to gate enrollment on (icpId; when omitted, the org's most-used ICP is auto-selected as the default gate, if the org has one). Returns the new campaign in draft status, with no bound lists and no graph yet — follow with campaign_bind_list for an audience and campaign_set_graph for a sequence before campaign_publish. 404s (COMMON_NOT_FOUND) if channelAccountId is not a connected account in this org.

Pause an active campaign — { id }. Nothing new fires going forward; runs already in flight simply wait at their current step and resume where they left off if the campaign is reactivated later (nothing already sent is recalled). Idempotent: pausing an already-paused campaign is a no-op that returns the current campaign unchanged. Rejected (COMMON_VALIDATION_FAILED, 400) if the campaign is draft or archived — only an active campaign can be paused. 404s (COMMON_NOT_FOUND) if the campaign is not in this org.

Validate and publish the campaign's current draft graph — { id, migrateInFlight? }. Publish can FAIL validation: rather than stopping at the first problem, it returns published: false with EVERY validation error in errors (each { code, message, nodeKey?, condition?, params? }, codes like empty_graph, no_start, multiple_starts, unreachable, cycle, required_outcome_unwired, unknown_variable, template_body_required, ai_instruction_required, invalid_max_length, fixed_text_too_long) so every problem can be fixed in one pass — this is a value in the RESULT BODY, never a thrown exception. On success, published: true and version carries the freshly-published graph (version is null while published is false). migrateInFlight (default false) additionally repoints the campaign's in-flight runs onto the freshly published version by node key, reporting migration: { migrated, canceled } — a run parked on a node the new version removed is canceled instead of migrated. 404s (COMMON_NOT_FOUND) if the campaign is not in this org.

Read ONE sequence run in full — { runId } (an id from campaign_runs). Returns the lead, the sending seat, the run's position in the sequence, and steps — the node-by-node timeline with each step's status and timing. When a send is drafted or already out, body carries the actual text, with reasoning and evidence for an AI draft and verification when the send was checked. This is how you answer "what did we actually say to this person, and what happened next" — campaign_runs gives you the page, this gives you the story. 404s (COMMON_NOT_FOUND) if the run is not in this org, or belongs to another member's sending seat.

Read the campaign's sequence runs — one per enrolled lead — keyset-paginated: { id, bucket?, cursor?, limit? }. bucket filters to one activity tab: waiting (still in flight), didnt_accept (invite never accepted), didnt_answer (message sent, no reply), replied, interested (a reply classified interested — a subset of replied), skipped (an Enrollment Filter / ICP / exclusion verdict, or an operator stop), error (failed); omit bucket to list every run. Each run carries the lead (display card), currentNodeKey/currentNodeType/currentNodeRunStatus while live, nextActionAt for the live waiting bucket, outcome once terminal, the errorCode when outcome is failed, and skip/replyClassification detail where relevant. Page with the returned cursor. 404s (COMMON_NOT_FOUND) if the campaign is not in this org.

Replace the campaign's editable DRAFT sequence graph in full — { id, graph: { nodes, edges, startNodeKey } }. This OVERWRITES the whole draft; it is not a patch. Each node is { key, type, config, uiPosition: { x, y } } — key is a stable string you choose (referenced by edges), type is one of send_invite | wait_connection_accepted | delay | send_message | like_last_post | visit_profile | comment_on_post | research (wait_reply isn't supported yet), and config is that type's own shape (e.g. { note, draftMode?, instruction?, language? } for send_invite — its length budget is computed automatically, not a config field — { days } for delay, { timeoutDays } for wait_connection_accepted). Edges are { fromKey, toKey, condition }, referencing node keys by the outcome they route. The wiring rules, per node type — the legal outcome-edge conditions (flagging which MUST route onward) and the legal successor types:

Everything about how a campaign is performing, in one read — { id, days? }. summary is the live execution snapshot: { status, audienceSize, enrolledCount, queuedCount, activeRunCount, usedToday, dailyEnrollmentLimit, sendWindow, currentVersion } — currentVersion is null until campaign_publish has run, which is the thing to check before campaign_activate. funnel is the outcome funnel (enrolled → invited → accepted → messaged → replied → interested). buckets is the count per activity bucket — where everyone currently stands (queued, waiting, didnt_accept, didnt_answer, replied, interested, skipped, error) — the same vocabulary campaign_runs' bucket filter takes, so a non-zero count tells you which filter is worth reading. activity is the daily series, and is returned ONLY if you pass days (7 or 30). Its days are the campaign's own local days in its send-window timezone, not UTC. 404s (COMMON_NOT_FOUND) if the campaign is not in this org.

Edit a campaign's configuration — any subset of { id, name, sendWindow, autonomy, dailyEnrollmentLimit, enrollmentFilter, icpId, stopOnReply, leadListIds }. Omitted fields are left unchanged. sendWindow is { days: [1..7 as Mon..Sun], startHour, endHour, timeZone } and is REQUIRED before campaign_activate will run. autonomy is 'autopilot' (sends fire on schedule) or 'manual' (each send waits for approval). icpId gates enrollment on an ICP (null clears the gate — enroll everyone). stopOnReply: false measures replies without halting the run. leadListIds binds lead lists to the audience and is **additive**: lists already bound stay bound, and a bound list you leave out is NOT unbound — unbinding would cancel every in-flight run drawn from that list, so it is not something this tool can do by omission. Returns the updated campaign. 404s (COMMON_NOT_FOUND) if the campaign, a lead list, or icpId is not in this org.

List LinkedIn conversations, newest activity first. Defaults to YOUR OWN seat — pass scope: "org" for every seat in the organization, or seatId for one teammate. Filter with unread (only chats with unread messages), leadsOnly (only chats linked to a lead), and q (case-insensitive match on the counterpart's NAME — this does not search message text). Each chat carries id (for chat_messages), counterpartName, leadId when linked, unreadCount, and lastMessagePreview. Page with the returned cursor. Reading is free: it is never metered and never counts against a sending limit.

Read one conversation's messages, **newest first**, by the chatId from chat_list. Each message has direction ("in" = from them, "out" = from you), body, sentAt, and any attachments; a message the counterpart deleted comes back with deleted: true and an empty body. Page further back in history with the returned cursor. Opening a thread may pull fresh messages from LinkedIn, which is not metered and does not count against any sending limit.

Send an InMail (optional subject). Address the recipient with target — { type, value } (url, public_identifier, or provider_id). Requires the Sales Navigator seat and a live InMail credit; refused fast otherwise. Optionally attach files with attachments — each item is EITHER { fileName, mimeType, dataBase64 } (the file bytes inline as base64; at most ONE such file per send, ≤3 MB — if you cannot produce base64 reliably, e.g. as a chat agent without direct file access, use url instead) OR { fileName, mimeType, url } (an https url we fetch server-side; up to 5 of these, ≤15 MB each) — never both on the same item. This contacts a real person and consumes a credit — there is no human approval step.

Send a LinkedIn connection invitation (optional note) through the caller's account. Address the recipient with target — { type, value } (url, public_identifier, or provider_id). Refused if already connected/invited; deduped per recipient; capped daily. This contacts a real person — there is no human approval step.

Retract a message YOU sent, addressed by the chatId and message id that chat_messages returns. Only outbound messages can be deleted — an inbound one is a not-found. Removes it from LinkedIn and marks it deleted in the Kairon conversation. Idempotent. Metered.

Replace the body of a message YOU sent, addressed by the chatId and message id that chat_messages returns. Only outbound messages can be edited — an inbound one is a not-found. LinkedIn allows this on Classic messages within about an hour of sending; later attempts are refused by LinkedIn. Updates the conversation in Kairon as well as on LinkedIn. Metered.

Send a LinkedIn direct message. Pass EXACTLY ONE of: chatId — reply in an existing conversation (the id from chat_list), which is what you want whenever the conversation already exists; or target — { type, value } (url, public_identifier, or provider_id) to open a NEW conversation, allowed only with a 1st-degree connection or an open profile. Optionally attach files with attachments — { fileName, mimeType, dataBase64 }, up to 5 files, each ~3 MB. Returns chatId and, for a reply, the persisted message (with the id message_edit and message_delete take). This contacts a real person — there is no human approval step. Metered against the seat's daily send limit.

The users you may act as, each with their organization and whether they have a LinkedIn account connected — you can only act as someone who does. Staff see every tenant; an org owner or admin sees the members of the organizations they govern. search matches name, email or organization.

Act as another user from now on: every later call runs as them, on their LinkedIn seat and their organization. Staff may act as anyone; an org owner or admin only as a member of their own organization, never one ranking above them. user is an email address or a user id. Calling it again switches target. The mode belongs to this client alone and ends 60 minutes after your last call, or on impersonate_stop.

Go back to being yourself. Succeeds whether or not you were acting as anyone, and reports which it was.

Takes no parameters.