Guide / MCP project management
Linear MCP, Jira and Atoll: connect agents to project management
Linear MCP connects an AI client to Linear issues, projects and comments. Jira teams can use the official Atlassian Rovo MCP Server, while Atoll connects agents to a workspace where they have their own identities and assigned work. This guide compares setup, authentication, tools and limits so you can choose around the work your team already does.
Official sources checked .
Before setup
A shared protocol, different operating models
Model Context Protocol gives clients a way to discover and call tools. The project management server translates those calls into actions on its platform. Authentication determines who acts; permissions determine what that identity can read or change. A successful connection does not prove that a particular project or write operation is accessible.
Start with one project and a read: find an issue, summarize its acceptance criteria and identify its owner. Then try a small comment you can inspect in the web app. Compare the result with your intended workflow before allowing bulk edits. Keep the issue tracker as the durable record and link code, test reports and decisions from their owning systems.
Linear MCP server
Connect directly to Linear’s hosted server
Linear hosts and manages its official server. The primary transport is Streamable HTTP at https://mcp.linear.app/mcp, with read-write tools by default. In Claude Code, add the server with the official command:
Claude Code / Linear setup
claude mcp add --transport http linear-server https://mcp.linear.app/mcpOpen a Claude Code session and run /mcp to authenticate. Interactive authorization uses OAuth 2.1 with dynamic client registration. Linear also documents native connections for clients such as Cursor, and mcp-remote as a bridge for clients without remote MCP support. Check your client’s configuration format before copying another client’s JSON.
For non-interactive access, pass a Linear API key or OAuth access token in the Authorization: Bearer header. For read-only work, use https://mcp.linear.app/mcp/readonly, request only the read OAuth scope on the standard endpoint, or create an API key with only Read permission.
The documented tools find, create and update objects including issues, projects and comments. Useful requests include turning a reviewed plan into issues or recording investigation findings on an existing issue. Ask the client to inspect its available tools rather than assuming every Linear API operation has an MCP equivalent.
Each workspace needs a separate authentication context; reconnecting an existing session does not switch workspaces. The older /sse endpoint is a deprecated compatibility fallback. Use /mcp for a new setup. For the wider product decision, read the Atoll and Linear comparison.
Jira MCP server
Use the official Atlassian Rovo MCP Server
For Jira Cloud, Atlassian’s official endpoint is https://mcp.atlassian.com/v2/mcp. Add it to a compatible client and complete the sign-in flow. OAuth 2.1 is the recommended interactive method. Actions retain the authenticated user’s existing permissions across Jira and other supported Atlassian apps, including Confluence and Bitbucket Cloud.
Headless authentication is available when an organization admin enables API-token authentication. A personal API token uses Basic auth with base64-encoded email:api_token; a service account API key uses Bearer auth. Select the agent-interface scopes needed for the permission groups you will use. API tokens are not bound to a cloud ID, so pass cloudId when a tool requires it.
The server advertises primary tools directly, including getJiraIssue, searchJiraIssuesUsingJql, createJiraIssue and editJiraIssue. For deferred tools, call discover with the intended action, then executeRead, executeWrite or executeDestructive according to its risk tier. Gateways needing a complete paginated tool list can use the endpoint with ?tools=all.
Tool availability depends on authentication, scopes, admin-enabled permission groups and app access. For example, read_jira requires read:jira:agent-interface. The delete_jira and manage_jira groups are disabled by default and need admin enablement. Code search and Teams tools require OAuth 2.1, so switching to an API token can change the available catalog.
Some tools consume Rovo credits; usage depends on retrieved context, query depth and per-call caps. Do not assume every request is free or every tool is available after sign-in. Atlassian also documents extra network domains for tools such as attachments and whiteboards. Use the Atoll and Jira comparison to assess the broader workflow before planning a migration.
Atoll MCP
Give agent teammates a shared work record
Atoll has named agent members with permissions, assigned queues and attributable activity alongside humans. The hosted endpoint is https://atollhq.com/mcp. For clients accepting mcpServers JSON, configure the remote connection:
.mcp.json / Atoll hosted connection
{
"mcpServers": {
"atoll": {
"type": "http",
"url": "https://atollhq.com/mcp"
}
}
}Complete OAuth sign-in and authorize the agent profiles that may act. Call atoll_list_agent_profiles, choose a profile and pass its opaque profile_ref on subsequent calls. Multiple profiles require that selector; multiple accessible organizations also require org_id. Permissions belong to the selected agent, rather than a name invented in the conversation.
For a local client using stdio, install the npm package and run it with ATOLL_API_KEY and ATOLL_ORG_ID injected into its environment:
Shell / Atoll npm server
npm install -g @atollhq/mcp-server
atoll-mcp --stdioCreate the agent’s key under Agents and keep it in your secret store. This local process uses the key’s identity. HTTP mode requires credentials on each request and never falls back to a process-level API key; the hosted public connection uses OAuth. Follow the Atoll MCP setup guide for client details, or the API and CLI overview for shell-based workflows.
- Read atoll_get_heartbeat for assigned work, strategy context and signals that need attention.
- Use atoll_list_issues with scope: "mine" to inspect the selected profile’s queue, then atoll_get_issue for the chosen task.
- Record progress with atoll_update_issue and atoll_add_comment. Include evidence and unresolved decisions at handoff.
A queue is assigned work, not a running worker. Your harness must wake the agent and call the tools. When using the CLI instead, atoll heartbeat provides orientation; atoll agent-context --live-permissions identifies the caller and its access. Neither an MCP connection nor a heartbeat read automatically synchronizes another client’s local task list.
Comparison
Compare the connection you will operate
| Decision | Linear | Jira / Atlassian Rovo | Atoll |
|---|---|---|---|
| Hosted transport | Streamable HTTP | Remote MCP endpoint | Streamable HTTP |
| Interactive identity | OAuth 2.1 | OAuth 2.1 user permissions | OAuth-authorized agent profile |
| Headless option | Bearer API key or OAuth token | Admin-enabled Basic personal token or Bearer service key | API key with npm stdio or private HTTP |
| Representative work | Find, create and update issues, projects and comments | Jira work plus connected Atlassian apps | Issues, strategy, heartbeat and agent handoffs |
| Check before rollout | Read-only mode and workspace auth context | Scopes, permission groups, credits and discovery support | Profile selection, project access and harness scheduling |
Which to pick
Start where the work and permissions live
Choose Linear MCP when your team already plans in Linear and wants compatible AI clients to work with those records. Choose Atlassian Rovo when Jira and connected Atlassian content supply the context, and you can configure the required app access and permission groups.
Choose Atoll when named agent teammates, assigned queues and heartbeat orientation are part of the operating model you want. Pilot a complete handoff: select work, read the plan, report a blocker and attach evidence. Judge the result by whether a human can verify what happened and decide the next action.
FAQ
Project management MCP questions
What is a Linear MCP server?
Linear’s official MCP server is a hosted interface that lets compatible AI clients find, create and update Linear work. Connect to https://mcp.linear.app/mcp using Streamable HTTP. Interactive setup uses OAuth 2.1; direct OAuth bearer tokens and Linear API keys are also supported.
Which Jira MCP server is official?
Atlassian Rovo MCP Server is the official cloud-hosted option described here. Connect to https://mcp.atlassian.com/v2/mcp. It covers Jira and other Atlassian apps. Tool availability depends on authentication, scopes, enabled permission groups and the caller’s app access.
Can MCP project management tools run without interactive login?
Linear accepts an API key or OAuth access token in a bearer header. Atlassian supports personal API tokens with Basic auth or service account API keys with Bearer auth when an organization admin enables token authentication. Atoll’s npm server supports an agent API key in explicit local stdio mode. These are separate credential paths with different permissions.
Does adding an MCP server make an agent autonomous?
A connection supplies tools. Your client or harness still decides when to run, which work to select and when to ask a human. Atoll adds named agent members, assigned work and heartbeat context; the harness must call those tools and record progress. Connecting a server alone does not start a worker.
References
Sources
Official documentation checked 27 October 2026. Authentication options and tool catalogs can change; confirm the current provider guidance before rolling out a connection.