MCP Connectors – TextCortex Help Center

What Are Connectors?

Connectors are the integrations that give your AI access to the tools and data your team already uses.

Without them, the AI works from whatever you paste into the chat. With them, it reads and writes directly from the source.

How Connectors Work

When you activate a connector, you give the AI permission to interact with a specific platform during a session.

Connectors started as read tools: pull a ticket, fetch a contact, check a deal status. They're becoming much more than that. Write tools mean your AI can now take action directly inside the platforms where your team works: draft a support ticket, create a task, update a deal stage, post a summary, log a note. No tab-switching, no copy-pasting.

Connectors vs. Knowledge Base Integrations

Some platforms, like Microsoft SharePoint, can be added to TextCortex in 2 ways: as a knowledge base integration or as a connector. They sound similar, but they do different jobs.

Knowledge base integration

Copies (syncs) content from the platform into a TextCortex knowledge base ahead of time.
The AI searches that synced copy and answers with references to the original documents.
Read-only. It can't change anything in the source platform.
Content is only as fresh as the last sync (manual, or daily automatic sync).
Best for: asking questions across large sets of documents, like policies, handbooks or project archives.

Connector (MCP)

Nothing is copied in advance. The AI connects to the platform live, at the moment you ask.
Always works with the current version of a file, record or ticket.
Can read and, with write tools, take action: create, upload, update, move.
Acts through the account you connected, so it only reaches what that account can reach.
You control every tool individually (see Managing Connectors & Permissions below).
Best for: live lookups and getting work done inside the platform.

In short: a knowledge base is something the AI searches. A connector is something the AI uses.

You can use both for the same platform. For SharePoint, a common setup is a knowledge base for company-wide Q&A plus the connector for working with specific files. See Microsoft SharePoint Integration for both setups.

Connectors and Skills

A connector opens a door. A skill decides what to do once the door is open.

The same Asana connector can power a sprint setup skill, a task summary skill, and a deadline tracker skill. The connector handles access. The skill handles logic. See What Are Skills? for how workflows get encoded.

Common Connectors and What They Enable

Google Drive

A Google Drive connector lets the AI search, read, upload, move, copy, and organize files directly from chat. Ask it to find last quarter's proposal, pull a specific doc into context before drafting a response, or organize files into folders. No opening Drive in a separate tab.

Microsoft SharePoint

A SharePoint connector lets the AI search sites, browse document libraries and lists, and read Word, PDF and Excel files live. With write tools, it can create folders, upload documents, copy and move files, add list items and update specific Excel cells. A reporting skill uses it to save a finished report straight into the right team folder. (SharePoint can also be synced as a knowledge base. See Connectors vs. Knowledge Base Integrations above for the difference.)

Microsoft Outlook

An Outlook connector gives the AI access to your inbox and calendar. It can read and send emails, and search inbox history. A meeting prep skill uses it to pull relevant email threads and upcoming events before a call. A follow-up skill uses it to draft and send replies directly.

Microsoft Teams

A Teams connector lets the AI search chat and channel messages, pull meeting transcripts, schedule Teams meetings and post to chats or channels. A meeting recap skill uses it to fetch the transcript, pull out action items and post the summary to the team's channel.

Notion

A Notion connector lets the AI read and write pages, databases, and docs. A project kickoff skill uses it to create briefs and populate templates. A knowledge base skill uses it to search internal docs and pull relevant context before drafting a reply.

Slack

A Slack connector lets the AI post updates, summaries, and alerts to specific channels. Most workflow skills end with a Slack step: the project kickoff summary goes to #projects, the weekly support digest goes to #support-ops, the SEO report goes to #marketing.

Jira

A Jira connector lets the AI read and act on issues, epics, and sprints. An engineering triage skill uses it to pull open bugs by severity, flag anything unassigned past a threshold, and surface blockers before the standup.

Zendesk

A Zendesk connector gives the AI access to tickets, customer history, and tags. A support skill uses it to pull open tickets, filter by urgency, and draft replies based on issue type.

Klaviyo

A Klaviyo connector lets the AI read segments, check campaign performance, and draft email flows. An email marketing skill uses it to pull a target segment, check suppression lists, and produce a sequence with copy, subject lines, and send timing.

HubSpot / Salesforce

A CRM connector is the backbone of most sales skills. It lets the AI check whether a company is already a customer, find open deals, look up contact ownership, and filter prospecting lists before anything reaches an SDR. Sending outbound to an existing customer is the kind of mistake that only happens once, and a CRM connector prevents it.

Apollo

An Apollo connector gives the AI access to contact and company data for outbound prospecting. A sales skill uses it to search for accounts by industry and size, enrich contacts with email and role data, and score leads by intent signals before cross-checking them against the CRM.

Chaining Multiple Connectors

Skills often chain 2 or 3 connectors in sequence. A project kickoff skill might write a brief to Notion, build the task tree in Asana, and post a summary to Slack.

A prospecting skill might search Apollo, filter through HubSpot, and export to a formatted list.

The skill defines the sequence. Each connector just does its part when called.

Advanced Connector Use

As skills get more specialized, connectors do heavier work.

An SEO audit skill might call an analytics platform for domain metrics, pull ranking data for 200 keywords, identify traffic-losing pages over 90 days, and write a full HTML report, all in one session.

A content research skill might query a CMS across 1,000+ posts, run 5 parallel keyword research queries, and check competitor rankings, chaining a CMS connector, an SEO connector, and a file output.

The connectors are the same ones you already activated. The skill just asks more of them.

Custom MCP Server

If we don't support a tool your team relies on, you don't have to wait for us to build it.

Our Custom MCP Connector framework lets you connect any third-party system's MCP server to TextCortex yourself.

If your industry runs on a niche CRM, an internal tool, or anything not on our standard list, you can wire it in directly if the tool supports an MCP server. Once connected, it works like any other MCP connector: your skills can call it, read from it, and write to it.

Where to add a custom server

From a chat: open Tools -> Connectors and add a custom server. It's available to you as soon as it's connected.
Tenant admins: in the tenant admin settings, go to Authentication & Access -> MCP Connectors and click Add Custom Server in the Shared custom servers section. That's also where you share it with your workspace (see below).

Custom servers are only available if your workspace admin has enabled custom MCP connectors.

What you need

Setting one up takes 3 things:

Display name: What the connector will be called inside TextCortex. Pick something your team will recognize at a glance, like "Internal CRM" or "Warehouse API".
Server URL: The MCP endpoint for the third-party tool. You'll find this in that tool's MCP documentation, usually listed under something like "MCP server" or "agent integration". Copy the exact path: a server can respond on one path (for example /mcp) and return an error on another, so a bare domain often won't work.
Authentication method: How TextCortex proves it's allowed to connect. Choose one of the 3 options below.

Choosing an authentication method

No authentication: For public MCP servers that don't need any credentials.
API key: Paste the API key or token that authorizes TextCortex to connect, along with the header it should be sent in (the tool's docs tell you which one, often Authorization). Generate the key from the third-party platform's developer or API settings.
Sign in with the provider (OAuth): There's no key to copy. When you save, TextCortex sends you to the provider's sign-in and consent screen. Approve access and you're brought back to TextCortex with the connector added. This only works if the provider lets new apps register automatically for OAuth. If sign-in fails, check the provider's MCP docs or use an API key instead.

For no-auth and API key servers, you can click Test before saving to make sure the URL responds as an MCP server. The test is optional. It isn't shown for OAuth servers, because there's nothing to test with until you've signed in.

Sharing a custom server with your workspace

Tenant admins can share a custom server so that other people see it as a connector card in the Connectors menu, under the Custom section, just like the built-in connectors.

In Shared custom servers, pick the server from Select a custom server.
Choose who gets access: everyone in the workspace, or only the people on an allowlist.
Set the approval mode and switch the server on.
Save your changes.

When you share, TextCortex first checks that the server responds. If it doesn't, you'll get a warning and can either cancel or choose Share anyway. This keeps a broken server from reaching your whole workspace by accident.

What people need to do once a server is shared depends on how it authenticates:

No authentication: They switch it on. There's nothing else to set up.
OAuth: Each person clicks Connect and signs in with their own account on the provider. The admin's sign-in is never passed on to anyone. Reconnect and Disconnect only affect that person's own connection.
API key: Each person can add their own key. Admins can also offer their own key to the people they share with. The first time you do this, a consent dialog explains that anyone with access to the connector can use the key, that their usage counts against that key's permissions, quota and billing on the third-party platform, and that TextCortex never shows them the key. People can then choose between the admin-provided key and their own. Admins can revoke or rotate a shared key at any time.

Sharing a server never shares your key automatically. Your key only becomes available to others after you've confirmed that consent dialog.

Using custom servers in agents

Personal and shared custom servers also appear in the agent editor's Integrate catalog, so you can attach them to an agent.

Sharing that agent doesn't share access to the server: anyone using the agent still needs access to the connector and their own connection (or a shared key).

A note on security: API keys and tokens from the third-party system work like passwords. Never share them, don't paste them into chat, and store them in a secure location (a password manager or secrets vault, not a spreadsheet). If a key is compromised, revoke it immediately from the third-party platform, generate a new one and replace it in TextCortex. TextCortex stores saved keys encrypted and only ever shows them masked, so it can't display or recover a key for you. If the tool supports it, signing in with the provider means you never have to handle a key at all. Admins: if you share your own key, everyone you share it with uses your account's permissions and quota on the third-party platform.

Managing Connectors & Permissions

Connector permissions can be configured within the conversation by going to Tools -> Connectors, or through your TextCortex settings in the Integrations permissions section.

Each connector guides you through the login process (typically using OAuth or an API key) and defines what the AI is allowed to access.

Control each tool

Open a connector and go to the Tools tab to see every tool it offers. Tools that only read are listed separately from tools that make changes.

On/off switch: turn any single tool on or off. The AI can't use a tool that's switched off.
Ask First: for tools that make changes, the AI shows you what it's about to do and waits for your approval.
Instant: the AI runs that tool without asking. Best kept for low-risk, repetitive actions.
Use in this chat: switch the whole connector on or off for the current conversation.

For example, the Microsoft SharePoint connector has 14 tools, and 6 of them make changes (like Upload File and Move Item). You can leave all the read tools on, keep Ask First on for moves and uploads, and switch off anything your team doesn't need.

Once active, any skill that references it works automatically.

To see how connectors get wired into workflows, read Skills.