GTM Brigade

npx add-mcp 'https://api.gtmbrigade.com/mcp'

What it is

The GTM Brigade connector is a remote Model Context Protocol (MCP) server. One server, the same tools, whichever assistant you connect it to. Once connected, your assistant can answer questions like “how did my posts do this week?” and carry out work like “draft a post about our pricing change and schedule it for Tuesday” or “add these five profiles to the watchlist.”

It exposes 102 tools across your whole GTM Brigade account: 43 read your data and 59 change something. 43 are flagged destructive and 26 reach outside GTM Brigade: they publish on LinkedIn under your own name, send an email, write into your CRM or outreach tool, fetch a web page or image, or read public LinkedIn data through a data provider. Every tool that changes something is flagged as such, so your assistant asks you before it runs. ChatGPT shows 92 of the 102: plans, invoices, connecting an outreach provider, and building, validating, publishing, resuming and test-running workflows stay off ChatGPT. Of those 92, 39 are read-only, 53 change something, 38 are flagged destructive and 23 reach outside GTM Brigade.

For worked examples of what people actually ask it, see what you can do with the connector: real sentences, the tools each one fires, and which of them change your account.

Setting the account up, from the chat

This used to be the part you had to do in a browser. It is not any more. A brand-new user can create the workspace itself and work through setup in the same conversation:

• Create the workspace. The only write in the connector that runs when you belong to no workspace at all. It previews the name and the URL slug first and writes nothing until you confirm.
• Ask what is still unset. One read walks LinkedIn connection, company context, ICPs, leads and lists, watchlist, knowledge base, personal style, teammates, CRM, Company Page, targets, feed keywords and onboarding state, and hands back a link per gap.
• Fill the company context. Four free-text fields go into every AI prompt GTM Brigade runs. While they are empty, every generated post and comment is written with no idea what your company does.
• Seed the knowledge base from URLs. Hand it your own site, a pricing page, a case study. Our server fetches each page, extracts the readable article and queues it for indexing.
• Get ICPs proposed from your own website and commit the ones you pick, the same step the signup wizard runs.
• Find real people to engage with. Pull the reactors and commenters off a post or an account, or have candidates proposed from your own LinkedIn activity.
• Connect an outreach provider with an API key you paste, and write your personal style, including drafting it from the posts and comments of yours we already hold, which is the single highest-value field to fill.
• Mark onboarding complete, so the app stops pulling you back into the signup wizard. It reports what is still unset before it agrees you are done.

What still needs a browser

Setup is not entirely chat-shaped, and it is worth knowing which steps will send you to a window before you start:

• Signing up for GTM Brigade, and the connector's own OAuth consent screen. You need an account before there is anything to connect.
• Connecting your personal LinkedIn account. Nothing publishes or gets measured until that is done, and it is an OAuth handshake in a browser.
• Connecting a CRM or a Company Page, and Slack or Teams notifications, which additionally need per-member consent. These are all OAuth handshakes. The connector can tell you whether each one is live; it cannot perform the handshake.
• Stripe checkout and plan changes. Billing stays readable only.
• Uploading a file. Web pages by URL and text notes go into the knowledge base from chat; a PDF or a deck off your disk is still an upload in the app.

Add it to Claude

1. In Claude, open Settings → Connectors → Add custom connector.
2. Enter the server URL:

   https://api.gtmbrigade.com/mcp
   

3. Claude opens a browser window to sign in. Log in with your GTM Brigade account, review the permissions listed on the consent screen, and click Allow.
4. That's it. The tools appear in Claude.

Add it to ChatGPT

1. In ChatGPT, open Settings → Connectors and add a custom connector (this needs developer mode on a plan that allows custom connectors).
2. Enter the same server URL:

   https://api.gtmbrigade.com/mcp
   

3. Sign in on the GTM Brigade consent screen and click Allow, exactly as above.

The same URL and the same account work in both. ChatGPT's research surface looks for two specifically-named tools, search and fetch, and the connector serves them, so your leads, posts, drafts and knowledge base are citable there. Note that ChatGPT's deep-research mode only uses the reading tools and will not run the ones that write.

Add it to Gemini or Perplexity

Both connect to the same URL, with one difference: they may ask you for a client ID and client secret instead of registering themselves. Gemini Enterprise always does (its admin console has no automatic registration), and Perplexity falls back to it. Ask us and we will issue a pair for your organization.

• Gemini Enterprise: add a custom MCP server in the admin console, enter https://api.gtmbrigade.com/mcp and the credentials we issue. If the console shows a PKCE checkbox, tick it: this server requires PKCE and the connection will not complete without it.
• Perplexity: add a custom connector under your account or organization settings (Pro, Max, Business or Enterprise). Enter https://api.gtmbrigade.com/mcp, and the credentials if it asks for them.

Neither Google nor Perplexity runs a connector directory, so there is nothing to browse for. You add the URL yourself.

Authorization uses OAuth 2.1 with PKCE in every client. GTM Brigade never sees your Claude, ChatGPT, Gemini or Perplexity account, and none of them ever sees your GTM Brigade password. You sign in on our own consent page.

What it can read

43 read-only tools, in thirteen groups:

Group	Tools	What it covers
Setup and integrations	3	The workspaces you belong to, a workspace’s setup status with what is still unset and a link per gap, and what is connected: your outreach providers, your organization’s CRM, and Slack and Teams notifications.
Your analytics and your team’s	6	Your posts, comments, reactions, impressions, reach, followers, activity score and leaderboard rank, plus the team leaderboard, org-wide overview and Company Page metrics for owners.
Leads and lead lists	4	Search your lead database, open a single lead with its enrichment and engagement history, list your lead lists, and follow the progress of enrichment running in the background.
Watchlist, feed and comment suggestions	7	The profiles your team monitors with their activity stats, the posts they published, a single feed post in full, your saved feed views, the brief behind comment suggestions, and the AI comment suggestions waiting on them.
Content and drafts	4	Your posts and drafts with their status and performance, the personal writing style GTM Brigade learned from you, and a draft of style instructions written from it (nothing is saved until you set it).
Knowledge base	2	List your organization’s knowledge-base assets, and search them semantically. Your private personal assets stay private to you.
ICPs	2	Your ideal-customer profiles, their signals, and how many leads match each one.
Comment automation	2	Your automation settings and the queue of drafts waiting for approval or scheduled to go out.
Feed keywords	1	The keywords you subscribe to for keyword-matched posts in your feed.
Billing, plans and consumption	3	Your plan, credit balances, consumption breakdown by user and feature, invoices, and the available plans.
Search and fetch	2	One search across your leads, posts, drafts and knowledge base, and a fetch that opens any result in full. These carry the names ChatGPT expects for its research surface, and work the same way in every client.
Team and targets	2	Your organization’s member roster (owners only) and the activity targets set for the team.
Workflows	5	Your automation workflows, one workflow’s draft and published version, the result of a single run, the catalog of node types the engine offers, and a validation pass that tells you why a draft is not ready to publish.

What it can change

59 tools write. They do exactly what the equivalent button in the app does, with the same permission checks, the same validation and the same credit accounting, driven by Claude instead of a click. Anything that spends credits is quoted to you first: Claude tells you what it will cost and what your organization has left, and only goes ahead once you say so.

Group	Tools	What it covers
Setup and integrations	5	Create a workspace and become its owner, fill the four company-context fields that get injected into every AI prompt, connect Aimfox, HeyReach, lemlist or SendPilot with an API key you paste, run your LinkedIn scorecard, and mark onboarding complete. Creating a workspace previews first and writes nothing until you confirm, and it is capped at three a day.
Content	9	Generate a post with AI, create and edit drafts (including attaching an image by URL), schedule a draft for automatic publishing to LinkedIn, publish a draft to LinkedIn immediately, unschedule a scheduled draft, delete a draft, fetch your own LinkedIn posts and comments so the AI can learn your voice, and save your personal style.
Leads	7	Add profiles by LinkedIn URL, pull every reactor and commenter off a LinkedIn post or account into your organization, get people proposed from your own LinkedIn activity, enrich leads you already have, assign leads to teammates, delete leads (owners only), and push a lead out to your connected HubSpot, Salesforce, Pipedrive or Attio (owners only). Anything that costs credits is quoted first: Claude shows the exact cost and what your balance will be afterwards, and can only spend against a quote you approved.
Lead lists	6	Create, rename and delete lists, add and remove leads from them, and flag leads as ready for outreach.
Watchlist and engagement	5	Remove profiles from the watchlist, generate a comment with AI, submit a generated comment for review, post a comment on LinkedIn, and react to a post on LinkedIn.
Feed state, views and keywords	6	Claim, unclaim or mark posts as seen for your team, create, edit and delete saved feed views, and subscribe or unsubscribe from feed keywords.
Comment automation	3	Change your automation settings (daily cap, approval mode, quiet hours), and approve or reject a queued draft.
Knowledge base	3	Add a text note, ingest public web pages by URL so our server fetches and indexes them, and delete an asset.
ICPs	7	Have ICPs proposed from your own company website and commit the ones you pick, then create, edit and delete ICPs by hand, recompute their matches, and apply them to a lead list.
Team and targets	2	Invite a person to your organization by email (owners only), and set the team’s activity targets.
Workflows	6	Create and edit an automation workflow as a draft, run the draft once against a subject you name, publish the draft so it starts firing on its own, pause a live workflow, and resume a paused one on the version it was already running.

The tools that leave GTM Brigade

Most tools only touch data inside your account. 26 of the 102 do not (23 of the 92 on ChatGPT). Each one is flagged in the protocol, so your assistant asks first. They fall into six kinds, and the first four are visible to other people or land in someone else's system, so they cannot be taken back from here.

• Public on LinkedIn, under your own name. Publishing a post, posting a comment and reacting go out immediately, using the LinkedIn account you connected in the app, with no undo from GTM Brigade: taking something down is a manual job on LinkedIn. Scheduling a post, approving a comment-automation draft and turning on comment automation do the same later, when our publisher runs. Publishing, resuming or test-running a workflow (in Claude) arms an automation that then acts on its own, repeatedly, as you (posting, commenting, sending DMs, writing to your CRM) with no prompt each time it fires. Test-running is not a rehearsal: it runs the draft for real. Publishing previews the difference against what is running today before it commits.
• Email. Inviting a team member sends an invitation email to the address you name.
• Your CRM or outreach tool. Pushing a lead writes that lead's details (name, email, company, job title, LinkedIn URL) into your connected HubSpot, Salesforce, Pipedrive or Attio. It goes out under your organization's shared CRM connection rather than your personal one, so it is restricted to owners. It matches on the LinkedIn profile URL, then the stored CRM id, then the email address, so pushing the same lead twice updates the existing record, and a lead with neither a LinkedIn profile nor an email address is refused. In Salesforce it creates a Lead, not a Contact. Flagging leads as ready for outreach marks them for your outreach tool.
• The public leaderboard. Completing setup adds you to the public GTM Brigade leaderboard when your LinkedIn is connected and audited.
• Fetching a web page, an image address or a web search. Adding a web page to the knowledge base, attaching an image by URL to a draft, and proposing ICPs from your company website make our server fetch that address, so those sites see a request from us. What comes back is untrusted third-party text and is handled as data, never as instructions. Generating a post may use Google Search to ground the text.
• Public LinkedIn data, read through a data provider. Adding leads, discovering leads from a post or account, proposing leads, enriching leads, fetching your own LinkedIn content for your personal style, drafting a comment (the post's most relevant public comments, on posts with 5 or more) and the daily search behind a feed keyword all read public LinkedIn data through a data provider, and the ones that cost credits are quoted first.

Connecting an outreach provider validates the API key you paste against Aimfox, HeyReach, lemlist or SendPilot before it is stored encrypted against your own user. It is never echoed back, logged, or shown again. Reading your integrations re-checks each stored key the same way. Both only touch your own private account with that provider, so they are not counted as reaching outside.

What “destructive” means here

43 tools are flagged destructive (38 of the 92 that ChatGPT shows), which is what makes your assistant warn you before running them. Every tool that removes, replaces or publishes something carries the flag, and none of them is a hard wipe. They fall into four kinds:

• Removes something. Deleting a draft, lead, lead list, ICP, saved feed view or knowledge-base asset; taking a profile off the watchlist (its collected posts and assignees go with it, unused comment drafts are discarded, and the person stays in your lead database); taking leads off a list; unsubscribing from a keyword. Lead and lead-list deletion is the app’s soft delete, recoverable only by us, on request. Default lists, such as the watchlist, cannot be deleted at all.
• Overwrites or replaces something. Updating a draft, ICP, lead list, feed view or workflow; every tool that sets something (company context, personal style, targets, post state, comment automation); assigning leads; adding leads to a list; applying proposed ICPs, which replaces your ICP criteria and lead-scoring rules, so a configuration you tuned by hand is gone (the first call previews what would be replaced and changes nothing); refreshing proposed leads or suggested ICPs, which replaces the stored list, and a refresh that fails replaces it with an empty one; refreshing your LinkedIn audit; connecting an outreach provider, which replaces the key you stored before.
• Acts in the world and cannot be taken back. Publishing or scheduling a post, posting a comment, reacting, approving or rejecting an automation draft, inviting a member, pushing a lead to your CRM, flagging leads as ready for outreach, and completing setup, which lists you on the public leaderboard. Publishing, resuming or test-running a workflow (in Claude) belongs here too: it runs for real, as you, with no pretend mode. Enrichment and lead discovery spend credits and update leads you already have.
• Cancels something that was going to happen. Unscheduling a post, and pausing a live workflow.

Access & permissions

The connector respects the exact same permissions as the app. Who you are and what you may do are worked out on our side from the account you signed in with, never from anything Claude sends us. If Claude asks for another organization's data, the request simply does not resolve.

Every member can read and act on their own work. Owner-only tools (team leaderboard, team overview, Company Page analytics, member roster, invitations, lead deletion, ICP deletion) give a member a plain “requires owner access” message, never another person's data. Everything is scoped to the organizations you belong to.

Safety controls

• Granular scopes. Access is split across 19 permissions: analytics, billing, leads, content, engagement, the watchlist, the knowledge base, the team, ICPs, settings and workflows, most of them split into reading and writing. Each tool checks its own scope before doing anything, so a connection granted part of the list reaches exactly that part. Drafting a workflow and letting one act are deliberately two different grants, so you can hand your assistant the drafting half without the acting half.
• Per-user, per-organization isolation. Every query is scoped to your organization on our side. Your private personal knowledge-base assets stay isolated to you, including when Claude generates content from them.
• Hourly budgets on outward actions. Public LinkedIn actions, invitations, AI generations, CRM pushes and web-page ingestion each have their own per-hour cap per user, and workspace creation is capped per day. A runaway agent loop hits the cap and stops with a clear message instead of spamming anyone.
• Claude asks first. No tool that changes something is marked read-only, so your client prompts you before running one, and warns harder on the destructive and outward-facing ones.
• Your normal audit trail. Writes run through the same code paths as the app, so they land in the same records (activity logs, credit ledger and audit log), attributed to you.
• Yours to revoke. Remove the connector in Claude at any time. Access tokens are short-lived and bound to this connector and your account.

What it deliberately cannot do

Some things were left out on purpose, because doing them safely needs more than a wrapper:

• Repost. GTM Brigade has no repost action anywhere, and the connector does not invent one.
• Message anyone on LinkedIn. There is no tool to send a DM, an InMail, or a connection request. The connector can comment and react in public; it cannot slide into anyone's inbox.
• Change roles or remove members. It can invite, and that is all: a decision we made deliberately, not a gap. Lockout and billing consequences stay with a human.
• Buy anything. No checkout, no plan changes, no payment details. Billing is read-only: consumption, invoices, plans.
• Merge or hard-delete leads. Lead deletion is the app's recoverable soft delete; merging duplicates is not exposed at all.
• Upload files. Text notes and public web pages by URL go into the knowledge base from chat. A file off your disk is still an upload in the app: there is no way to move a binary over this transport.
• Run an OAuth handshake for you. LinkedIn, the CRM, the Company Page, Slack and Teams all connect in a browser. The connector reports whether each one is live and, for outreach providers, accepts a pasted API key. That is the whole of it.
• Touch platform administration. Our internal staff tooling is not on the connector at all.

Privacy

See our privacy policy for what the connector accesses, what it changes, and who your data is shared with when you use it.

Support

Questions or issues? Email ovidiu.ionita@gtmbrigade.com.

Server URL

https://api.gtmbrigade.com/mcp

The same URL in Claude, ChatGPT, Gemini and Perplexity.