What Tin MCP server does
Tin connects a coding agent to a hosted marketing workspace. The agent can work with a Tin project, use the project’s files and retained context, and start marketing workflows from the catalog. The repository describes 60 available workflows spanning organic search, content, social publishing, outreach and listings, product QA, paid advertising, competitor research, and creative tasks. Availability depends on each workflow’s inputs, connected services, and project settings.
Work can continue beyond one agent session. Saved workflows can have schedules, while Tin manages timers, retries, waits, and approval steps. Research reports, a project wiki, and other context live in a git repository associated with the project. For work outside the catalog, project.task creates a separate Codex task that can ask questions, pause, resume, and propose file changes for approval.
How it works
The Tin MCP server is accessed through the hosted endpoint https://app.tin.computer/mcp. A coding agent connects over MCP, authenticates through the browser, and then shares the project with Tin according to the permissions you approve. Tin can use information from the local repository during the setup conversation, while project-specific files and decisions remain available to later runs.
Workflows describe their required inputs, outputs, and review needs. Some run on demand; others can be saved with a schedule. Content and other deliverables do not automatically go live merely because a plan contains dates. Approved content can remain in project Files, or GitHub can deliver it as a pull request or a commit to main, depending on the project’s delivery setting. Technical fixes still require the operator to merge them.
Setup and configuration
Connect the Tin MCP server from the project directory used by your coding agent. Claude Code can add the HTTP connection with claude mcp add --transport http tin https://app.tin.computer/mcp; authentication is then started with /mcp. Codex uses codex mcp add tin --url https://app.tin.computer/mcp, followed by browser login when prompted. Cursor requires an MCP entry in .cursor/mcp.json with the same URL, after which its MCP controls can complete authentication.
After connecting, start a new agent session if the tools are not visible. Calling list_projects checks whether the connection is working. Connections belong to individual projects, and Tin checks the access required by a workflow before it runs. Examples in the documentation include read-only Google Search Console access, a GitHub App for repository reads and pull requests, and other service connections used by particular workflows.
Tools and capabilities
Tin exposes capabilities for:
- Listing and working with Tin projects.
- Running catalog workflows for search, content, social, outreach, QA, advertising, competitors, and creative work.
- Creating separate project tasks with
project.task.
- Maintaining project files, reports, research, and wiki context.
- Reviewing generated work, requesting revisions, and approving campaigns, posts, ads, or pages.
- Delivering approved content through GitHub or leaving Markdown in project Files.
- Using local
.agents/skills/SKILL.md material when a workflow declares the relevant skill and that material has been saved to the Tin project.
The workflow registry is the source of truth for each workflow’s inputs and requirements. Tin can also report that a planned content item is already covered or lacks sufficient evidence instead of generating another version.
Limitations and notes
The Tin MCP server is hosted rather than a local repository process, so authentication and service connections are handled through Tin’s web application. A local skill file is not automatically available to hosted runs; the relevant material must be saved to the Tin project. Workflow availability varies by configuration and connected services.
Content-plan dates are editorial targets, not guaranteed publication times. Tin does not automatically draft and publish an entire six-month calendar. Approved GitHub content may use a pull request or a commit-to-main setting, but technical changes wait for user approval. The README also marks workflow evaluation and growth-experiment tracking as forthcoming rather than current capabilities.