How MCP Connectors Let Claude Use a Scheduler

Claude has no social accounts of its own: it reaches a scheduler through a connector, which is a remote MCP server that the scheduler’s vendor hosts. Once you add the connector and sign in, Claude can call the tools that server exposes (list your channels, save a draft, schedule a post), and the scheduler, not Claude, publishes at the set time. According to Anthropic’s help centre, checked October 12, 2026, the Free, Pro, Max, Team and Enterprise plans can each add a custom connector, and the Free plan is limited to one.

This post explains the mechanism and gives the steps as Buffer documents them today. The Postnavo setup is on posting to social media from Claude with Postnavo, which is launching soon.

What a connector is, and where MCP fits#

MCP, the Model Context Protocol, is “an open-source standard for connecting AI applications to external systems” (MCP project). A connector is how Claude uses it: “Connectors let Claude access your apps and services, retrieve your data, and take actions within connected services.” (Claude Help Center)

A custom connector is a remote MCP server you add yourself. MCP servers used to run only on the user’s own computer, and developers can now host them on the internet (Claude Help Center). A scheduler is that remote case, “a cloud service you sign into” in the words of Anthropic’s guide to connector types. The local case is a desktop extension, which installs an MCP server on your computer (Claude Help Center).

MCP is one of three ways a scheduler can open itself to an agent. What an agentic social media scheduler is lists it beside a CLI and an API.

Host, client and server#

In MCP’s terms Claude is the host, the AI application that connects to one or more MCP servers, and a server is “A program that provides context to MCP clients” (MCP architecture overview). For a scheduler, the vendor runs the server. The same page says remote servers accept bearer tokens, API keys and custom headers, and that MCP recommends OAuth.

Tools are what Claude calls#

A server offers tools, which the architecture page defines as “Executable functions that AI applications can invoke to perform actions”. The model decides when to call one (MCP server concepts). The client lists a server’s tools with tools/list, then executes them.

The vendor decides what the tools do, and that can include creating, changing or deleting data in the connected application (Claude Help Center). So Claude can do only what the scheduler’s server exposes. Postiz’s documentation states one gap: “There is no getComments or replyToComment exposed via MCP.” (Postiz docs)

What happens when you ask Claude to schedule a post#

Say you type this:

Write a LinkedIn post announcing our new pricing page and schedule it for Thursday at 10am. Also make a shorter version for X.

The list below is what Anthropic’s, Buffer’s and Postiz’s documentation provides for that request. Each item names its page, and none of it comes from a run.

1. Tools Claude can invoke. “Remote MCP servers give Claude tools it can invoke during your conversation.” (Claude Help Center)
2. A tool to read the account and the channels. In Buffer’s MCP guide, get_account returns the account’s timezone and local time, which the guide says an assistant should use to turn “tomorrow at 5pm” into a real timestamp, and list_channels lists the connected social accounts. Postiz’s equivalent is integrationList (Postiz docs).
3. A tool that returns a network’s rules, in Postiz. Postiz documents a tool, integrationSchema, from which the agent “learns X’s character limits, settings, and rules”. Whether Claude checks a draft’s length without such a tool is not documented by Anthropic, so paste the draft into a character counter yourself.
4. Approval before a write, in most clients. Of the tools that change your data, Buffer’s guide says: “Most MCP clients ask you to approve those before they run.” That is not a guarantee: Claude has an “Allow always” choice, and during Research it “can invoke tools from your connectors automatically without further approval” (Claude Help Center).
5. The write tool. Buffer’s create_post “Schedules or publishes a post on one channel”, so with Buffer the prompt above means two calls. Its mode is addToQueue (the default), shareNow, shareNext or customScheduled.
6. The saved record. In Buffer’s words, the write tools “hand back the record they saved”.

The scheduler publishes, not Claude#

According to both vendors, Claude does not have to stay open. Buffer’s product page says to “let Buffer handle the publishing” (Buffer MCP), and Postiz’s Claude page says: “Posts publish at the scheduled time even after you close the chat.” (Postiz)

Each network’s media rules still apply. Buffer’s create_post reference lists, as a minimum, “Instagram and TikTok: an image or a video.” The rest is in what Instagram requires of any tool.

Which Claude plans and apps can use a connector#

You do not need a paid plan. This table is from Anthropic’s custom connectors page, checked October 12, 2026.

Where a remote connector works, according to Anthropic’s two help pages (connectors, custom connectors) and the Claude Code documentation, checked October 12, 2026:

Adding a custom connector from the mobile app is not documented: Anthropic says installing connectors on mobile is “currently in beta” and names desktop and web as the primary path (Claude Help Center).

For a connector added in Claude on the web, Desktop, Cowork or mobile, the connection runs “from Anthropic’s cloud infrastructure, rather than from your local device” (Claude Help Center). So a scheduler you host behind a firewall or a VPN will not connect unless you allowlist Anthropic’s IP ranges in your firewall, which the same page describes. The Claude Code documentation does not state this for a server added with claude mcp add. It describes a sign-in in your browser whose callback goes to a port on your own machine (Claude Code docs).

How to connect a scheduler to Claude: Buffer’s documented steps#

One scheduler that documents this route today is Buffer. It documents a hosted MCP server and a Claude setup. These are Buffer’s steps as published, checked on October 12, 2026. They were not run by us.

The server is at https://mcp.buffer.com/mcp (Buffer’s MCP guide). Sign-in from Claude is by OAuth, “No API key needed.” (Buffer’s Claude guide), and “MCP access is included on every plan, including Free.” (Buffer MCP)

In Claude on the web or the desktop app#

From Buffer’s Claude integration page:

1. Open Claude in your browser or in the desktop app.
2. Navigate to “Customize”, click “Connectors”, then click “Add custom connector”.
3. Fill in the form with the name Buffer and the remote MCP server URL https://mcp.buffer.com/mcp, then click “Add”.
4. Click “Connect”, sign in to Buffer if prompted, and approve access.

Buffer’s developer guide gives another route in place of steps 2 and 3: open Connectors and search for “Buffer” in the connector list (Buffer’s Claude guide). Anthropic’s public connector listing at claude.com/connectors, checked October 12, 2026, shows neither Buffer nor Postiz (Claude Marketplace), so use the custom connector route. The in-app directory at claude.ai/directory returned an access check and was not read.

Anthropic’s own flow has more screens than Buffer’s step 3. After the URL, its page shows a Continue step and an authentication screen before the final Add (custom connectors page). Anthropic’s pages also word the opening button two ways: + Add, then Add custom connector (custom connectors page), or Add, then Custom, then Web (connectors page). If one does not match your screen, look for the other.

Anthropic says the OAuth sign-in lets Claude act for you “without Claude ever seeing your actual password” (Claude Help Center). After it, you switch the connector on for a chat from the “+” button at the lower left of the chat, under Connectors.

In Claude Code#

claude mcp add --transport http buffer https://mcp.buffer.com/mcp

Then, per Buffer, start a session, run /mcp, select the Buffer server and sign in. This matches Anthropic’s Claude Code documentation, whose syntax is claude mcp add --transport http <name> <url>. Two more facts from that page:

• A connector added in claude.ai is “automatically available in Claude Code” when you are logged in with a claude.ai account.
• A hand-written JSON entry needs "type": "http" beside its url: “A JSON entry that has a url but no type is a configuration error”.

Postnavo’s pages on scheduling posts from Claude Code and on a CLI and agent skill describe what it is launching with for coding agents.

Postiz documents the same route#

Postiz documents an MCP server with 13 tools (Postiz docs). Its Claude page says: “Paste the Postiz MCP link in Claude settings > Connectors > Add custom connector.” and “No API key needed.” (Postiz)

Postiz’s addresses differ by how you sign in. The docs give https://mcp.postiz.com/mcp, which takes an API key as a Bearer token, a second form with the key in the address, and for a self-hosted install /mcp-oauth, which they call “OAuth-protected” (Postiz docs). The Claude page gives https://mcp.postiz.com/mcp-oauth-dynamic, which takes an OAuth sign-in and no key.

For Claude Code, Postiz’s Claude page gives the command in Anthropic’s syntax: claude mcp add --transport http postiz https://mcp.postiz.com/mcp-oauth-dynamic. The docs’ setup page writes it with a --url flag and an API key, a form Anthropic’s page does not show.

Both schedulers, from each vendor’s own pages, checked October 12, 2026. They are compared from documentation, not used. Postnavo is not in the table because it is not released.

Keep a person between Claude and the publish button#

The MCP specification says: “For trust & safety and security, there SHOULD always be a human in the loop with the ability to deny tool invocations.” (MCP specification)

On any plan, Anthropic’s advice is to review each approval request, to click “Allow always” only for “a server and tool that you trust to run unsupervised”, and to use the “Search and tools” menu to disable tools you do not want Claude to invoke (Claude Help Center). A setting for each tool of Always allow, Needs approval or Blocked is documented for Team and Enterprise owners, under Customize > Connectors, where it applies to the whole organization (Claude Help Center). Buffer is blunter: “Leave the approval prompts on. create_post can publish immediately, and delete_post cannot be undone.” (Buffer’s MCP guide)

Buffer’s product page says something else: “Everything goes through your Buffer queue, so nothing publishes without you seeing it first.” (Buffer MCP) Read the two together: the tool can publish at once if asked to, so the approval prompt and the draft option are what keep a person in the loop. Ask for a draft by name. Buffer has saveToDraft, and Postiz’s type takes draft, schedule or now (Postiz docs).

Before you connect, know three things:

• The access is wide. Buffer: “The connection covers your whole account.” Of its API keys it says: “There is no per-organization scoping at this time” (Buffer docs).
• Claude may bring it into a chat without you naming it. “Claude can bring up a connected app on its own when it fits what you’re doing.” (Claude Help Center)
• You can undo it. Disconnect the connector in Claude’s settings (Claude Help Center), or revoke access “from Settings → API in Buffer” (Buffer MCP).

Add only the vendor’s own address, because Anthropic warns that malicious MCP servers may carry hidden instructions (Claude Help Center). Buffer states: “The only Buffer MCP endpoint is https://mcp.buffer.com/mcp.” (Buffer’s MCP guide)

A human-in-the-loop approval checklist covers the full workflow.

When it does not work#

This list is ordered by what is cheapest to check first. It is compiled from the vendors’ documentation, not from support cases.

1. The connector is off for this chat. Click “+” in the lower left of the chat, hover over Connectors and toggle it on (Claude Help Center). In Claude Desktop that menu also shows each server’s tools (Claude Help Center).
2. Tool access is set to “On demand”. Then “Connectors aren’t loaded until Claude searches for the right one based on your request.” Click “+”, hover over Connectors, then Tool access, and switch to Always available, which Anthropic recommends when “A specific connector needs to work every time without fail”. The default is Auto (Claude Help Center).
3. The sign-in failed or lapsed. “If authentication fails, try disconnecting and reconnecting from Customize > Connectors.” (Claude Help Center) How long a sign-in lasts is not documented.
4. You are on Team or Enterprise and an owner has not added it, or has restricted it. An Owner or Primary Owner has to enable connectors for the organization first. An owner can also set a tool to Blocked for the whole organization, and members cannot override that (Claude Help Center).
5. Claude Code does not list it, or lists it as hidden. Check the type field of a hand-written entry. Then check your login: “Connectors from claude.ai are fetched only when your active authentication method is a claude.ai subscription login.”, so they are not loaded while ANTHROPIC_API_KEY or a third-party provider is active. A server you added in Claude Code takes precedence over a claude.ai connector at the same URL, and /mcp then lists the connector as hidden. If /mcp shows session token rejected, run /login, then reconnect the connector from /mcp (Claude Code docs).
6. The server cannot be reached. This applies only to a server you host yourself. A custom connector added in Claude on the web, Desktop, Cowork or mobile is reached from Anthropic’s cloud, not from your machine. If the server is behind a firewall or on a private network, allowlist Anthropic’s IP ranges in the firewall (connectors page, custom connectors page). The Claude Code docs do not say this for a server added with claude mcp add.
7. The tool call returns 429. In Buffer, “Past the limit the call fails with 429 Too Many Requests”, and the error carries a Retry-After value in seconds (Buffer’s MCP guide). The figures are on Buffer’s rate limits page, checked October 12, 2026: on the Free plan each client gets 100 requests in 15 minutes, 250 in 24 hours and 3,000 in 30 days.
8. The post was created but a network rejected it. That is the network’s side. Ask Claude to read the failure: in Buffer, “A post with the error status carries the failure message.” (Buffer’s MCP guide) One documented reason a scheduler’s link to a network lapses: on LinkedIn, by default, “access tokens are valid for 60 days and programmatic refresh tokens are valid for a year”, and the member must reauthorize when refresh tokens expire (LinkedIn’s developer docs). The causes on each network are in why a scheduled social media post does not publish.

FAQ#

Which Claude plans can add a custom connector?#

Free, Pro, Max, Team and Enterprise, with one custom connector on Free (Claude Help Center): the table is under which Claude plans and apps can use a connector.

What is the difference between a Claude connector and an MCP server?#

A connector is how Claude’s apps present a remote MCP server. An MCP server can also run on your own computer, as a desktop extension, and Anthropic says those are not available on web or mobile (Claude Help Center).

Do connectors work in the Claude mobile app?#

Yes, for one you already connected on the web or desktop, while adding a custom connector on mobile is not documented: see the table of apps.

Why does a connector show as connected when Claude cannot use its tools?#

Check the toggle for that chat, then the tool access mode, then on Team and Enterprise a tool an owner set to Blocked or restricted for the whole organization. In Claude Code, check the type field, the login method, a duplicate server at the same URL and a session token rejected state in /mcp. All are in the list above, and those are the causes on the pages cited here.

Does Claude have to stay open for a scheduled post to go out?#

No, according to Buffer’s and Postiz’s pages: see the scheduler publishes, not Claude.