Skip to content

Black Friday: up to 61% off our packs until November 30.

See the packs

Claude Code + ReShare MCP: Post an AI Character Video to 5 Networks with Human Approval

A step-by-step walkthrough of ReShare's 13 November 2026 test run, showing how to take one AI character video from a Claude Code conversation to five networks with a person approving every risky step.

This guide follows one Claude Code conversation that takes a short clip, swaps in an AI character, drafts captions for each network, previews the result and schedules it. A person approves every step that carries real risk.

What ReShare's 13 November 2026 test run actually showed

ReShare tested this workflow on 13 November 2026. In its write-up, one Claude Code conversation went from asking which ReShare account was connected to a video scheduled on TikTok, Instagram, YouTube, Facebook and Threads. Keep in mind this is ReShare's own test, written up on its blog, not an independent benchmark or an Optijara result.

One detail matters for planning: the character swap took longer than its first estimate. Step 3 has the numbers.

Why this works without custom integration code

Claude Code is only the client here. ReShare's MCP server exposes every tool the agent calls, from the first account check to the final publish, so you write no integration code. Anthropic introduced the Model Context Protocol as an open standard for connecting AI systems with data sources, replacing one-off integrations with a single protocol.

We have organised the walkthrough below into five steps:

  1. Connect ReShare to Claude Code with a dedicated key.
  2. Confirm the account, profiles and characters.
  3. Upload the clip and start the character swap.
  4. Write a caption brief for each network and run a preview.
  5. Publish with confirm: true and an idempotency key, then check what happened.

Before you start: ReShare plans, limits and what an agent can reach

Agents get the same plan limits as the ReShare app. On the Free plan you get 5 posts a month, 1 character and no AI swaps, and your agent receives the same limit message and upgrade link you would see. If a run hits a limit halfway through, that message is what comes back. Know your tier before the agent starts work, not mid-render.

Character swaps like the one in the test run need an AI plan. The table summarises the tiers ReShare lists, with Sharing plan prices rising with the number of profiles covered.

Plan typePriceWhat it covers
FreeFree5 posts a month, 1 character, no AI swaps
SharingFrom $9 a month ($90 a year) to $69 a monthUp to 10 profiles at the $69 tier
AI Starter$29 a month10 swaps, 720p, clips up to 10 seconds
AI Agency$499 a month150 swaps, 1080p, clips up to 15 seconds

The ReShare agents docs also list the limits that shape this workflow:

  • Uploads from an agent are capped at 95 MB per video.
  • X is not supported yet, so agents cannot post there.
  • Each key or connected app can make 120 requests a minute. Past that, the agent is asked to wait and try again.

Pre-flight checklist

  • [ ] You are on a plan that includes AI swaps if you want a character video.
  • [ ] The clip length fits your plan (up to 10 seconds on Starter, up to 15 seconds on Agency).
  • [ ] The file is within the agent upload cap listed above.
  • [ ] Every target network is already connected in the ReShare app.
  • [ ] X is not on your target list for this run.

Step 1: Connect the ReShare MCP server to Claude Code with a dedicated key

According to the reshareai/agents GitHub repo, everything goes through one remote MCP server using Streamable HTTP. The Claude Code MCP docs say HTTP servers are the recommended option for connecting to remote MCP servers. The same docs say the SSE transport is deprecated, so use HTTP.

First, create a dedicated API key in ReShare for this workflow alone. Check the ReShare agents docs for the permission options available on keys at the time you set this up, and give the key the narrowest access the workflow needs. Then add the server from your terminal:

  1. Put the key in an environment variable for the current session, for example export RESHARE_API_KEY=your-key.
  2. Add the server over HTTP, replacing RESHARE_MCP_URL with the endpoint address from the repo: claude mcp add --transport http reshare RESHARE_MCP_URL --header 'Authorization: Bearer '$RESHARE_API_KEY.
  3. Run claude mcp list to confirm the server is connected.

After adding the server, it is worth checking how the key ended up stored in your configuration. For this workflow we suggest local or user scope rather than project scope, so the key stays out of anything you might share with a team or commit to a repository. If you aren't sure what your local Claude Code files already contain, our guide to auditing Claude Code settings.local.json for secret leaks shows how to check them and rotate anything exposed.

Treat the key like a password

ReShare's agent safety guidance is direct: treat a key like a password, because it can post to your accounts. Keep it in your terminal or a password manager, never in a shared folder or a code repository. You can have 10 active keys, and when you reach 10, New key is turned off until you revoke one.

MCP authorization follows OAuth 2.1, and the MCP guidance says access tokens should be short-lived, because a stolen token works until it expires. We suggest applying the same thinking to API keys: rotate them on a regular schedule and revoke the ones nobody uses.

Step 2: Confirm the account before the agent touches anything

In the test run, the conversation opened by asking which ReShare account was connected. A prompt along these lines asks the agent to stay with questions only:

> Which ReShare account is connected? List my profiles and characters. Don't upload, render or post anything; stop and wait for my go-ahead.

Starting with questions alone can catch the wrong workspace, a network that was never connected in the ReShare app, or a plan without swaps, before any upload or render has started.

Step 3: Upload the clip and start the AI character swap

Have the agent upload the clip and confirm with you that the file arrived before anything else happens. The exact upload steps are set by ReShare, so check the ReShare agents docs for the current flow rather than assuming it stays the same.

Check the file against the upload cap in the limits list. If it is too big, trim or compress it before you hand it to the agent.

The agent then calls start_swap with the character you picked. When the agent called start_swap in the test, it got a video id back right away with an estimate of 480 seconds. The render took 650 seconds, about 11 minutes. An estimate is not a promise. Build in a buffer.

While the render runs, we suggest asking the agent to check status at a relaxed pace rather than in a tight loop. Each key or connected app can make 120 requests a minute, so hard polling eats into that budget for nothing.

Don't schedule anything until the agent confirms the render is done.

Step 4: Write one caption brief per network, then run a preview

Teams may find that each network rewards a different caption style, length and tone. A caption copied to every network won't fit any of them especially well. A workable brief covers:

  • Network: for example Instagram
  • Audience: who this post is for
  • Tone: friendly, expert or playful
  • Length: short or medium
  • Hashtags: which ones, or none
  • Call to action: what the viewer should do next
  • Scheduled time: date, time and time zone
  • Disclosure: how the caption says the video features an AI character

Repeat it for TikTok, YouTube, Facebook and Threads. The agent drafts the captions. You decide how the post is positioned on each network.

These videos feature an AI character, so add a line on disclosure to every brief, and if you plan to use C2PA metadata or watermarks, our AI content provenance playbook covers the options.

TikTok privacy is your call

Per the ReShare docs, the agent asks you who can view each TikTok post (everyone, friends, followers or only you) and never picks it for you. Answer that question yourself every time. TikTok has its own policy on top of that: its Content Posting API docs state that all content posted by unaudited clients is restricted to private viewing mode.

Review a full preview before publishing

Before anything is published, we suggest asking the agent for a preview that shows every caption, profile and scheduled time, and making sure nothing goes live at that stage. Read all of it. This is the last point where a person can catch a wrong caption, profile or time.

Step 5: Publish with confirm: true and an idempotency key, then verify

publish_post only works with confirm: true and a one-time key, so a retry never posts twice.

That one-time key is an idempotency key. It solves a very ordinary failure. The IETF Idempotency-Key draft describes it: a client sends a POST request, the request times out, and the client is left uncertain about the status of the resource. The draft also says that if a retry arrives while the original request is still being processed, the server should reply with HTTP 409.

After a timeout or a conflict response, we suggest having the agent wait and check the post status rather than resend with a new key. Changing the key would likely defeat the duplicate protection the key is there to provide.

Once publishing returns, we suggest having the agent look the posts up again to confirm the outcome. Don't rely on the first response alone. In the test run, the conversation ended with the video scheduled on all five networks.

The Four-Gate Agent Posting Routine: Read, Make, Words, Post

We suggest the Four-Gate Agent Posting Routine for teams adopting this workflow. We built it from the documented safeguards above, not from a client engagement. At the end of each gate a person has to say yes before the agent continues, and each gate has one concrete thing for that person to check.

Gate 1: Read

The agent reports which account is connected and lists the profiles and characters. Approval check: the approver confirms the workspace name, that every target profile appears, and that the chosen character is listed. Nothing is uploaded until they do.

Gate 2: Make

The agent uploads the clip and starts the swap. Approval check: the approver confirms the right file and character before the upload, then watches the finished render before anyone sets a publish time. Plan for the render to overrun its estimate, as it did in the test.

Gate 3: Words

The agent drafts captions from one brief per network. Approval check: the approver reads each caption against its brief, confirms the disclosure line is there, and sets TikTok privacy for every TikTok post.

Gate 4: Post

The agent shows the full preview, then publishes with confirm: true and a recorded idempotency key. Approval check: the approver signs off the preview's captions, profiles and scheduled times, then has the agent look the posts up again to confirm each one landed as planned.

Key hygiene

Use separate keys per purpose or per teammate within the 10-key limit. If you would rather the agent never handled a raw key at all, a credential proxy is one pattern to evaluate, and our playbook on the credential proxy pattern with Infisical Agent Vault explains how that approach works for AI agents.

Run log template

DateKey nameVideo idIdempotency keyNetworksApprover

With a log like this, you can trace any post back to the key that sent it, the render it came from and the person who approved it. So when someone asks why a video went out, you have the answer.

Common mistakes to avoid

  • Promising a publish time based on the render estimate. Wait until the agent confirms the render is done.
  • Uploading oversized files, or expecting the agent to post to X. Neither works today.
  • Pasting the API key into a project file or shared repo. Keep it in a password manager or as a terminal secret.
  • Adding the server with SSE. It is deprecated in Claude Code. Use HTTP.
  • Leaving TikTok privacy unchecked. Choose it yourself, and remember that unaudited clients are limited to private viewing.
  • Skipping the preview, or retrying publish_post with a new idempotency key after a timeout. Skipping the preview removes your review step. A new key throws away the duplicate protection.
  • Never revoking keys. Once you have 10 active keys, you can't create another.

Caveats and limits

Every timing and outcome in this guide comes from one vendor test run on 13 November 2026.

The plan limits, prices and supported networks here are taken from ReShare's docs and blog at the time of writing, and they may change. Check them before you commit budget or set a launch date. X is not available for agent posting yet, and your API client's audit status may limit TikTok visibility.

For teams running several agents at once, the per-key request limit and the cap of 10 active keys per account will start to matter.

The confirm and idempotency safeguards stop double posts. They do not stop bad posts. That job belongs to the person reviewing the preview.

Key Takeaways

  1. In ReShare's own test on 13 November 2026, one Claude Code conversation went from an account check to a video scheduled on five networks. It is a vendor test, not an independent benchmark.
  2. Add the ReShare MCP server to Claude Code over HTTP rather than deprecated SSE. We suggest local or user scope and a dedicated key with the narrowest access ReShare allows.
  3. Start with account, profile and character checks and confirm your plan limits. The Free plan has no AI swaps, so confirm your tier before the agent uploads anything.
  4. Leave a buffer in any publish time, because the documented render took longer than its estimate. Have the agent check status at a relaxed pace, not in a tight loop.
  5. Write one caption brief per network, choose TikTok privacy yourself, and read a full preview before anything goes live.
  6. publish_post needs confirm: true and a one-time idempotency key. After a timeout, we suggest checking the status rather than retrying with a new key.
  7. Use the Four-Gate Agent Posting Routine (Read, Make, Words, Post), with one concrete approval check per gate. Name keys by purpose, revoke unused ones within the 10-key limit, and keep a run log.

Conclusion

ReShare's test run shows that one Claude Code conversation can take a clip from upload to a scheduled post on TikTok, Instagram, YouTube, Facebook and Threads. Whether that is safe to run for a real brand depends on the sign-off at the preview step. Whoever approves that preview owns what goes out. The routine above makes sure that person sees the right account, the finished render and the final captions before saying yes. Start small: one short clip on one or two networks, with the run log filled in from the first post. That way you learn how renders, captions and approvals behave in your own accounts before adding more networks or more agents. If your team would like a second opinion on approval gates and key policies for agent publishing, Optijara can help you design them.

Frequently Asked Questions

Do I need a paid ReShare plan to make AI character videos with Claude Code?

Yes. The Free plan includes 5 posts a month, 1 character and no AI swaps. Character swaps need an AI plan. Starter costs $29 a month (10 swaps, 720p, clips up to 10 seconds) and Agency costs $499 a month (150 swaps, 1080p, clips up to 15 seconds). Sharing plans run from $9 a month ($90 a year) up to $69 a month for 10 profiles. Check current pricing before you commit.

Which transport should I use to add the ReShare MCP server to Claude Code?

Use HTTP. The reshareai/agents repo says everything goes through one remote MCP server using Streamable HTTP. The Claude Code MCP docs recommend HTTP servers for remote MCP connections and describe SSE as deprecated. Copy the endpoint address from the repo or the ReShare agents docs. We suggest local or user scope rather than project scope, so the key stays out of anything you share with a team or commit to a repository.

How long does a ReShare character swap take?

In ReShare's 13 November 2026 test run, start_swap returned a 480-second estimate and the swap finished in 650 seconds, about 11 minutes. That is one data point, so leave a buffer. Have the agent check status at a relaxed pace rather than in a tight loop, and don't schedule posts until it confirms the render is done.

Can a retry or timeout cause the same video to be posted twice?

Not if you use the safeguards properly. publish_post only works with confirm: true and a one-time key, so a retry never posts twice. Under the IETF Idempotency-Key draft, a retry that arrives while the first request is still being processed should get HTTP 409. After a timeout or a conflict response, we suggest waiting and checking the post status instead of sending a new key.

Can the agent post to X or choose TikTok privacy for me?

No to both. Agents can't post to X yet. For TikTok, the agent asks you who can view each post (everyone, friends, followers or only you) and never picks it for you. TikTok also restricts content posted by unaudited API clients to private viewing mode, so check what visibility you will actually get.

How should I store and manage ReShare API keys?

Treat each key like a password, because it can post to your accounts. Keep it in your terminal or a password manager, never in a shared folder or code repository. Each account can have 10 active keys, and New key is turned off at 10 until you revoke one. We suggest naming keys by purpose, rotating them on a schedule and revoking any you no longer use.

Sources

  1. https://www.reshare.ai/blog/make-videos-with-claude-post-everywhere
  2. https://www.reshare.ai/docs/agents
  3. https://www.reshare.ai/help/agents/agent-safety
  4. https://github.com/reshareai/agents
  5. https://code.claude.com/docs/en/mcp
  6. https://modelcontextprotocol.io/specification/2025-03-26/basic/transports
  7. https://modelcontextprotocol.io/docs/2026-07-28/tutorials/security/authorization
  8. https://www.anthropic.com/news/model-context-protocol
  9. https://datatracker.ietf.org/doc/html/draft-ietf-httpapi-idempotency-key-header
  10. https://developers.tiktok.com/docs/en/content-posting-api-get-started
  11. https://glama.ai/mcp/connectors/ai.reshare/reshare

Share this article

Hamza Diaz

Written by

Hamza Diaz

Hamza Diaz is the founder of Optijara, where he builds practical AI agents, automation systems, and Copilot workflows for service businesses. He writes about AI operations, agent strategy, and real-world implementation for teams that want usable systems instead of hype.

All articles

Next step

Have something to build?

Tell us what you need. We reply with a clear scope and a fixed quote.