Agent-first means the agent works from the real plan
Agent-first project management is not a separate board built only for bots. People still plan projects, set ownership and priorities, discuss work, and review progress in Scoped. An authorized agent can use that same current record instead of starting from a pasted brief that becomes stale as soon as the team changes it.
The shared plan is the stable layer across agent sessions. A person can hand work to an agent in one tool, another teammate can update the issue in Scoped, and a later session can begin from the resulting project state. Read the people-and-agents update walkthrough for a concrete write example.
MCP connects the client to structured project work
The Model Context Protocol, or MCP, is a standard way for an AI client to discover and call capabilities supplied by another application. Scoped exposes project-management context and actions through that interface. The Scoped Cloud endpoint is https://app.scoped.ai/mcp; a local or operator-run deployment uses its own application origin followed by /mcp.
MCP is the connection layer. Scoped remains the system that stores projects, issues, comments, ownership, priorities, relationships, and activity. The agent still runs in the client or environment you chose; Scoped does not host the model or start an agent run.
Authorization is tied to a person and workspace
A connection uses Scoped’s OAuth flow. The person connecting the client signs in, chooses a workspace, and approves the requested access. The resulting grant stays bound to that person, that workspace, the Scoped MCP resource, and its approved scopes. It does not turn the client into a workspace administrator or give it access to another workspace.
Scoped checks the current membership, workspace, requested resource, and granted scopes when the client accesses MCP. Agents use the application interface; they do not receive a PostgreSQL connection or a database credential. Switching workspaces in the browser does not retarget an existing grant.
Named setup paths exist for Codex and Claude Code
Scoped currently maintains explicit connection instructions for two MCP clients: Codex and Claude Code. Each guide uses that client’s own command syntax and the same Scoped OAuth consent boundary.
Another client may implement the same Streamable HTTP and OAuth standards. That protocol possibility is not the same as a named, maintained integration. The Scoped MCP integrations index lists only clients for which Scoped currently provides a setup path.
An authorized agent can read and return project work
The exact actions depend on the connection’s permissions. Within those limits, an agent can find relevant projects and issues, read current status and discussion, follow relationships between work, and bring progress back by updating an issue or adding a comment.
- Find context
- Search the workspace and retrieve the project or issue that matches the task.
- Read current work
- Use current fields, comments, and related records instead of a detached snapshot.
- Return progress
- Update permitted issue fields or add a comment so the result joins the shared record.
Accepted MCP changes keep the authorizing person and connection source in the activity record. See how an authorized agent update moves into Scoped.
Revocation stops future access, not completed work
You can inspect and revoke a connection. Revocation prevents future access through that workspace grant, but it does not erase comments or issue changes that were already accepted. Those changes remain part of the project record so the team can understand what happened.
MCP access is not permission to act outside Scoped. It does not let the product run the agent, approve external actions, or extend access beyond the authorizing person’s workspace rights. Read the full boundary on the what Scoped does for agents page.
Questions about Scoped MCP access
These answers describe the connection boundary that applies across the named client setup guides.
Is MCP the agent runtime?+
No. MCP is the connection between the agent client and Scoped. The model, prompt, tools, and agent process stay in the external environment you chose, while Scoped stores the project record and enforces access to it.
Does an MCP connection receive Scoped database credentials?+
No. The client calls Scoped’s application interface with an OAuth token. It does not receive a PostgreSQL connection string or direct database access.
Can one connection move between every workspace I can access?+
No. Consent binds a connection to the workspace selected during authorization. Switching workspaces in the browser does not retarget the existing grant; connect again and approve the other workspace separately.
What happens when I revoke an MCP connection?+
Revocation blocks future access through that workspace grant, including refresh. It does not erase comments, field changes, or activity that Scoped already accepted into the project record.
Connect project context without moving your agent
Start with the supported client guides, follow an agent update step by step, or return to the agent workflow overview. Create a workspace from the web app when you are ready to authorize a client against real project work.