All plugins

marsell

Run Marsell, an AI for marketing, sales and content creation, from Hermes through Marsell's hosted MCP server (Streamable HTTP at https://getmarsell.com/mcp, OAuth sign-in, no keys in config): read what Marsell planned, made and found for each project, answer the decisions waiting for the owner, read and update the brand, plan posts and other work, make pictures and films, run series, and check results and token balances. Bundles the marsell workflow skill. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. Disclosure: every tool call goes to Marsell's hosted server (getmarsell.com) and acts on the user's Marsell account with their current role; the tool set is served by that server at runtime and is not covered by the pinned sha. Tools that write change the owner's Marsell projects, and answering a waiting card in the owner's name can publish, send or spend as the chosen option says. Tools that make pictures, films and voice-overs, and some others such as research and page changes, spend Marsell tokens from the owner's plan; each says so in its description and asks the agent to give the price first. Ad spend needs the owner's approval in Marsell: an ad spends nothing until someone allowed to spend presses Start on its card there. Mission tools give the agent messages Marsell wrote for the owner's own Reddit, X or LinkedIn accounts, to send word for word from a signed-in browser, only where the owner said yes to that in Marsell. Upload tools give the agent a one-time address and a curl command that sends files from the user's machine, such as a logo or voice recordings, to Marsell. Sign-in runs the pinned mcp-remote@0.15.2 through npx (Node.js and npm required), which opens a browser and a local callback port the first time a tool needs the account and stores OAuth tokens under ~/.mcp-auth. The package has no telemetry and no self-updater. It marks the server untrusted, so Hermes asks before every tool the server does not mark read-only and refuses them in unattended runs.

MarsellToolsv1.0.0Updated 3 hours ago

About

Run Marsell, an AI for marketing, sales and content creation, from Hermes through Marsell's hosted MCP server (Streamable HTTP at https://getmarsell.com/mcp, OAuth sign-in, no keys in config): read what Marsell planned, made and found for each project, answer the decisions waiting for the owner, read and update the brand, plan posts and other work, make pictures and films, run series, and check results and token balances. Bundles the marsell workflow skill. Agent Plugins v1 package (plugin.json + mcp.json + skills/), configuration only. Disclosure: every tool call goes to Marsell's hosted server (getmarsell.com) and acts on the user's Marsell account with their current role; the tool set is served by that server at runtime and is not covered by the pinned sha. Tools that write change the owner's Marsell projects, and answering a waiting card in the owner's name can publish, send or spend as the chosen option says. Tools that make pictures, films and voice-overs, and some others such as research and page changes, spend Marsell tokens from the owner's plan; each says so in its description and asks the agent to give the price first. Ad spend needs the owner's approval in Marsell: an ad spends nothing until someone allowed to spend presses Start on its card there. Mission tools give the agent messages Marsell wrote for the owner's own Reddit, X or LinkedIn accounts, to send word for word from a signed-in browser, only where the owner said yes to that in Marsell. Upload tools give the agent a one-time address and a curl command that sends files from the user's machine, such as a logo or voice recordings, to Marsell. Sign-in runs the pinned mcp-remote@0.15.2 through npx (Node.js and npm required), which opens a browser and a local callback port the first time a tool needs the account and stores OAuth tokens under ~/.mcp-auth. The package has no telemetry and no self-updater. It marks the server untrusted, so Hermes asks before every tool the server does not mark read-only and refuses them in unattended runs.

Marsell MCP

Marsell is an AI for marketing, sales and content creation. Its hosted Model Context Protocol (MCP) server lets the AI you already use work with your Marsell projects.

This is Marsell's official public repository for connection instructions and MCP Registry metadata. The hosted backend is private and is not distributed in this repository.

Server URL: https://getmarsell.com/mcp
Transport: Streamable HTTP
Registry name: com.getmarsell/marsell, verified to getmarsell.com (it replaced io.github.Marsell-Official/marsell on 4 November 2026)

Full documentation · Try the interactive demo · Marsell · Privacy · Terms

Connect

Sign in with OAuth

In an MCP client that supports remote servers and OAuth, add https://getmarsell.com/mcp, choose OAuth authentication, sign in to Marsell and review the permissions before pressing Allow. Do not select No authentication. Client plans and administrator permissions can affect connector availability; see the connection guide.

For Claude Code:

claude mcp add --transport http marsell https://getmarsell.com/mcp

Then run /mcp, select marsell, and choose Authenticate.

Connect with an agent key

In Marsell, open a project, go to Agents → Your own AI, and choose Make a key. Choose Reads only, or Can act if your agent should also change things and spend tokens. For missions only, choose Make a missions key: it can only take and report missions. Copy the key when it is shown; it is only displayed once.

For Codex, set the environment variable in the shell that starts Codex:

export MARSELL_KEY='YOUR_MARSELL_AGENT_KEY'
codex mcp add marsell --url https://getmarsell.com/mcp --bearer-token-env-var MARSELL_KEY

On PowerShell, set the variable with $env:MARSELL_KEY = 'YOUR_MARSELL_AGENT_KEY', then run the same codex mcp add command.

Alternatively, configure Codex in ~/.codex/config.toml:

[mcp_servers.marsell]
url = "https://getmarsell.com/mcp"
bearer_token_env_var = "MARSELL_KEY"

Other clients can use the server URL with this HTTP header:

Authorization: Bearer YOUR_MARSELL_AGENT_KEY

Replace the placeholder with your own key through your client's secure credential settings. Never commit keys to Git or paste them into an issue.

Plugins and skills

This repository is also a plugin package. It connects the same hosted server and adds a skill, skills/marsell/SKILL.md, that tells your agent how to work with Marsell.

  • Any agent that reads skills: npx skills add Marsell-Official/mcp
  • VS Code: Install in VS Code
  • Gemini CLI: gemini extensions install https://github.com/Marsell-Official/mcp
  • Claude Code, Cursor and other plugin hosts: the manifests are in .claude-plugin/, .cursor-plugin/ and .mcp.json.

Hermes Agent

The root plugin.json and mcp.json are an Agent Plugins v1 package for Hermes Agent: configuration only.

Disclosure: every tool call goes to Marsell's hosted server at getmarsell.com, and the tools come from that server when Hermes connects, so the pinned commit doesn't fix them. Tools that write can change your Marsell projects. Tools that make pictures, films and voice-overs, and some others such as research and page changes, spend Marsell tokens from your plan; each says so before it runs. answer_decision answers a waiting decision in the owner's name, and the option chosen can publish a post, send a message or spend. The mission tools (missions_next, mission_done and the rest) give the agent messages to send word for word from the owner's own Reddit, X or LinkedIn account, and only where the owner said yes to that in Marsell. get_upload_link and voice_clone give the agent a curl command that uploads files from your computer to Marsell. Every ad spend still needs the owner's own approval in Marsell. Sign-in runs the pinned mcp-remote@0.15.2 through npx (Node.js and npm required): on the first tool call it opens your browser to Marsell and listens on a local port for the answer, then stores the OAuth tokens under ~/.mcp-auth. The package has no telemetry and no self-updater. It marks the server untrusted, so Hermes asks before every tool the server doesn't mark read-only and refuses them in unattended runs.

What your AI can do

  • Read projects, brand information, market research, people, analytics, token balances and decisions waiting for you.
  • With permission, answer decisions, update brand information, plan work and use Marsell's content, marketing and sales tools.
  • Create pictures with a token quote first, and work with Marsell's chat on content, canvases and videos.
  • Set up a business using the context and files your AI already has.
  • Pick up homework that needs your own AI's logins or files, and report its results to Marsell.
  • Send Marsell's messages on Reddit, X and LinkedIn from your own signed-in browser, one mission at a time at Marsell's pace, where you said yes to it in setup or Settings.

The tools available depend on your permissions. Your client can discover the current list through MCP tools/list; the full documentation describes the tools and their approval requirements.

Try asking:

What needs me on Marsell this week?

Read my brand and market, then tell me what we should work on next.

Check my Marsell homework and show me what is due before starting.

Homework does not create a background schedule by itself. If you want recurring checks, your AI client must support scheduling; it can report that schedule to Marsell with homework_schedule.

Permissions and account access

Reads only lets your AI inspect information. Can act also allows supported changes and actions that can spend tokens. Tool-specific approval requirements still apply, and Marsell checks your current account role on every call. Team and organisation access follows your Marsell permissions.

You can disconnect an OAuth app or revoke a key under Agents → Your own AI → Connected apps and keys. Account, plan and token requirements apply to the hosted service.

Troubleshooting

  • 401 Unauthorized: authenticate with OAuth or supply a valid agent key. The endpoint requires authentication.
  • Opening the URL in a browser fails: this is a Streamable HTTP MCP endpoint, not a web page. Use an MCP client.
  • A tool is missing or an action is denied: check the connection's scope, your current role and the tool's requirements.
  • A key stopped working: check whether it was revoked and whether the client can still read its environment variable.

OAuth discovery is available at protected resource metadata and authorization server metadata.

Repository and registry maintenance

server.json describes the hosted server for the official MCP Registry. It does not install or expose Marsell's backend.

Connection instructions and registry metadata are included in Marsell's MCP staleness checks. When the hosted interface changes, maintainers review this README and the manifest against the implementation. Registry metadata changes require validation with the official mcp-publisher and a new version before publication; changes in this GitHub repository alone do not update the registry.

Please report connection or documentation issues through this repository's issues, without credentials or private project data.

License

The documentation and metadata in this repository are licensed under MIT. That license does not cover the hosted Marsell service, its private backend, or Marsell's trademarks and brand assets.

README from the reviewed commit c7ed59d; it updates when the author re-pins.