Is a REST API enough, or do I need an MCP server?

An MCP server is easier for AI agents: the client discovers the tools and their inputs by itself, so nobody writes integration code. A REST API suits scripts with fixed logic, and it often covers more of the product. Flux has both, and they use the same API keys. Changes made through the Flux REST API are also recorded in the activity log, which MCP changes are not yet.

Can I give an AI agent read-only access to my project management tool?

On several tools, yes. Linear has a read-only endpoint, GitHub has a read-only path and header, Atlassian lets admins switch off write access (on Trello, Enterprise admins), and a Flux API key with only the Read scope can use the MCP read tools but none of the MCP write tools. Elsewhere, connect the agent with an account whose permissions are read-only.

Which project management MCP servers work for an unattended agent?

The ones that accept a key or token in a header, because an agent in CI or on a server cannot click through a browser sign-in. Linear, GitHub, monday.com and Flux document this. Atlassian does if an organization admin enables API tokens, and Notion does on its Business and Enterprise plans. Trello and ClickUp say they support OAuth only, and Asana's docs describe OAuth only.

Are remote MCP servers safe to connect?

They are as safe as the access you give them. An agent acts with the permissions of the account or key it uses, so give each agent its own credential with the narrowest scope that works, keep agents that only report on read-only access, and revoke credentials you no longer use. The MCP specification also asks host applications to get the user's consent before they invoke a tool.

Does Flux replace Jira or Linear?

It does not have to. If your team plans its work in Jira or Linear, their MCP servers let agents work there. Flux fits when you want one shared kanban board for people and agents, where agents connect with an API key and never take a seat, and you can have a person check agent work before it is done.

Project Management MCP Servers Compared (2026)

Which project management tools have an MCP server?

All eight trackers we checked run an official remote MCP server: Atlassian for Jira and Confluence, Linear, GitHub, Asana, Notion, Trello, monday.com and ClickUp. Flux has one too. They differ in five ways that matter for agents: sign-in, read-only options, the actions they cover, usage limits, and whether you can see what an agent changed.

An MCP server is the part of a product that AI agents use through the Model Context Protocol, an open standard for connecting AI applications to outside tools. A remote MCP server is hosted by the vendor. You give your MCP client, such as Claude Code, Cursor or Codex, a URL and a credential, and the agent can call the server's tools. There is nothing to install.

How we checked. We read each vendor's own documentation on 18 October 2026. Every page is listed in the sources. We did not sign in to the other vendors' products, so "not documented" means the pages we read do not say, not that the feature is missing. The Flux entry comes from the source code of the Flux MCP server. We build Flux, so weigh that entry accordingly.

Project management MCP servers at a glance

Last checked 18 October 2026. Endpoints are the vendors' hosted servers.

Tool	Endpoint	Agent sign-in	Read-only option	Status	What agents can change
Jira and Confluence (Atlassian)	`https://mcp.atlassian.com/v2/mcp`	OAuth 2.1; API token if an admin enables it	Admins can allow Read and block Write in Atlassian Administration	Generally available (Feb 2026)	Create and update work items and pages, change status, comment
Linear	`https://mcp.linear.app/mcp`	OAuth 2.1 or Linear API key	Read-only endpoint, read scope or read-only key	Not labelled	Find, create and update issues, projects and comments
GitHub	`https://api.githubcopilot.com/mcp/`	OAuth or personal access token	/readonly path or X-MCP-Readonly header	Generally available (Sep 2025)	Issues, pull requests, projects, repos and more, by toolset
Asana	`https://mcp.asana.com/v2/mcp`	OAuth	Not documented in the pages we read	Not labelled (beta server deprecated)	Create and manage tasks and projects
Notion	`https://mcp.notion.com/mcp`	OAuth; MCP token on Business and Enterprise	Not documented in the pages we read	Not labelled	Create and update pages and databases
Trello	`https://mcp.trello.com/v1`	OAuth 2.0 only	Enterprise admins can switch off write access	"Actively being developed"	Create, move and archive cards, comments, checklists; no permanent delete
monday.com	`https://mcp.monday.com/mcp`	OAuth or API token	--read-only flag (hosted server not stated)	Not labelled	60+ tools: create items, update columns, query boards
ClickUp	`https://mcp.clickup.com/mcp`	OAuth only	Not documented in the pages we read	Not labelled	Tasks, comments, time tracking, Docs; ClickUp pages disagree on task deletion
Flux	`https://flux.umin.ai/api/mcp`	Flux API key as a Bearer token (no OAuth)	Key with only the Read scope	Live, Free plan included	Create, update, move and reorder cards, add columns and comments; no delete

Where a vendor names the transport, it is Streamable HTTP, the remote transport defined in the MCP specification. It replaced an older HTTP and SSE transport, as the 2025-11-25 revision says. Linear and Notion keep an SSE endpoint only as a fallback, and monday.com no longer supports SSE.

What should you look for in a project management MCP server?

Check five things: how an agent signs in, whether you can keep it read-only, which actions the server covers, what agent usage counts against, and whether you can see what an agent changed.

OAuth or an API key?

Most of these servers expect a person to sign in. OAuth is the default for every vendor here except Flux, and the MCP specification's authorization rules for remote servers are built on OAuth 2.1. That works well for an agent you run in your own editor: you approve access once in the browser, and the agent acts as you.

An unattended agent is different. A CI job, a scheduled script or an agent on a server has nobody to click through a consent screen, so it needs a credential it can send in a header. Linear (API keys), GitHub (personal access tokens), monday.com (API tokens) and Flux (API keys) document this. Atlassian accepts API tokens if an organization admin turns them on, and Notion offers MCP tokens on its Business and Enterprise plans. Trello and ClickUp say they support OAuth only, and Asana's docs describe OAuth only.

Can you keep an agent read-only?

An agent that writes status reports or answers questions about the backlog does not need to change anything. Linear has a separate read-only endpoint, GitHub has a read-only path and header, Atlassian lets admins switch off write access in Atlassian Administration (on Trello, only Enterprise admins can), and a Flux key with only the Read scope can use the MCP read tools but none of the MCP write tools. monday.com documents a read-only flag for its server but does not say whether it applies to the hosted endpoint. For the rest, the docs we read describe no read-only mode, so the control you have is the permissions of the account the agent signs in with.

Which actions does it cover?

Having an MCP server does not mean an agent can do everything a person can in the app. Trello's server cannot permanently delete anything. ClickUp's pages disagree on deletion: its tool list has a Delete task tool, while its FAQ says no deletion tools have been added for now. Flux's cannot delete cards, assign people or add labels. Before you plan a workflow around a server, look at its tool list. Asana's docs make the same point: call tools/list to see the tools and parameters that are actually available.

What does agent usage count against?

Agents use these servers on somebody's plan, and vendors meter that in different ways. Atlassian charges MCP usage to Rovo credits. monday.com counts every tool call toward the account's daily API call limit. ClickUp publishes daily call limits, counted per workspace and per client, from 100 calls in 24 hours on Free Forever to 25,000 on Enterprise Plus when the workspace does not have the Everything AI add-on. In Flux, agents connected by API or MCP never take a seat; the pricing page has the plan details.

Can you see what an agent changed?

An agent works with the permissions of the credential it uses, as Atlassian says directly. Treat it like a colleague with that person's access. Atlassian's GA announcement also mentions MCP usage logs for audit and compliance. For the other vendors, the pages we read do not say how agent changes show up in history, so try it on a test project before you rely on it.

In Flux, comments an agent posts appear under the name of the person who created its API key, and cards it creates record that person as their creator. Changes made in the app, through the REST API and by the AI assistant are recorded in the activity log. Creating, editing, moving or deleting a card, and adding, editing or moving a column, can be undone while it is the latest change to that card or column. Changes made over MCP are not recorded yet.

How do the servers differ, tool by tool?

Mostly in how agents sign in and how far you can restrict them. Linear, GitHub, monday.com and Flux accept a key or token in a request header, while Trello, ClickUp and, going by its docs, Asana use OAuth only. Linear and GitHub also offer a dedicated read-only endpoint. Each section below covers one server from the table, in the same order.

Atlassian: Jira and Confluence

The Atlassian Rovo MCP Server has been generally available since 4 February 2026. Its current endpoint is https://mcp.atlassian.com/v2/mcp, and Atlassian says usage of the v1 endpoint moves to v2 on 1 March 2027. It covers Jira, Jira Service Management and Confluence, plus Bitbucket, Goals, Projects and Loom. Agents can search and summarize, and create and update work items and pages, for example moving an issue to a new status and adding a comment.

Sign-in is OAuth 2.1. API tokens work only if an organization admin enables them, either as a personal token with your email address or as a service account key, and some tools are not available with tokens. Atlassian recommends least privilege and reviewing high-impact changes, and its admin permissions let an organization allow Read and block Write. Comparing Jira and Flux for agent work? See Jira vs Flux and our Jira alternatives.

Linear

Linear launched its MCP server on 1 May 2025. The endpoint is https://mcp.linear.app/mcp, and it documents three ways to keep an agent read-only. You can sign in with OAuth 2.1 or send a Linear API key as a Bearer token. For read-only use, connect to https://mcp.linear.app/mcp/readonly, request only the read scope, or use an API key with only the Read permission. The tools find, create and update issues, projects and comments.

Linear's docs give the Claude Code command: claude mcp add --transport http linear-server https://mcp.linear.app/mcp, then /mcp to sign in. More on the two tools: Linear vs Flux and Linear alternatives.

GitHub

The remote GitHub MCP server at https://api.githubcopilot.com/mcp/ became generally available on 4 September 2025. It accepts OAuth or a personal access token. Its tools come in toolsets: the default set is context, repos, issues, pull_requests and users, and others include projects and actions. You can choose toolsets with a /x/{toolset} path or the X-MCP-Toolsets header, and make the server read-only with a /readonly path or the X-MCP-Readonly header.

Organizations on Copilot Business or Copilot Enterprise must enable the "MCP servers in Copilot" policy to use it from Copilot. GitHub also publishes a local version, which GitHub Enterprise Server needs. If you track agent work in GitHub Issues today, GitHub Issues vs Flux covers the trade-offs, and the GitHub integration guide shows how to create and move Flux cards from GitHub Actions with the REST API.

Asana

Asana's server is at https://mcp.asana.com/v2/mcp over Streamable HTTP, and its earlier beta server is deprecated. Clients must support OAuth, and some need a client ID and secret you create yourself. Agents can read tasks and projects, and create and manage them. On Enterprise+ plans, admins can allow or block each client. Asana's docs do not list the tools; they point you to tools/list. For a side-by-side of the two tools, see Asana vs Flux.

Notion

Notion MCP, at https://mcp.notion.com/mcp, lets agents search and read workspace content and create and update pages and databases. Most clients sign in with OAuth. On Business and Enterprise plans, a workspace owner can create an MCP token, which a client sends as a Bearer token with no browser sign-in. Guests cannot connect. Notion's older open-source server, notion-mcp-server, is no longer actively maintained. Notion's Claude Code command is claude mcp add --transport http notion https://mcp.notion.com/mcp. If your docs live in Notion and your tasks in Flux, see the Notion integration guide and Notion vs Flux.

Trello

Trello's official server is https://mcp.trello.com/v1. Atlassian documents it, and it works on all Trello plans. It uses OAuth 2.0 only: Trello says it does not currently support API tokens for MCP connections. Agents can search boards, create and move cards, mark cards done, archive cards and lists, and work with comments and checklists, but "destructive delete operations are not supported". On Trello Enterprise, admins can switch read or write access off in Atlassian Administration. Trello says the server is actively being developed, with more tools on the way. Trello vs Flux compares the two boards.

monday.com

monday.com hosts its Platform MCP at https://mcp.monday.com/mcp, with more than 60 tools for jobs such as creating items, updating columns and querying boards. It uses Streamable HTTP. You can sign in with OAuth, use your own OAuth app, or send an API token in an Authorization: Bearer header, and clients outside monday.com's featured list need an account admin to approve them. Every tool call runs as a GraphQL API request and counts toward the account's daily API call limit. The server's code, maintained by monday.com's AI team on GitHub, documents a --read-only flag. See also monday.com vs Flux.

ClickUp

ClickUp's server is https://mcp.clickup.com/mcp. It is OAuth only: ClickUp says personal API keys and access tokens won't work. Agents can create and update tasks, including custom fields and bulk changes, read and post comments, track time and work with Docs. Its tool list includes Delete task, but its FAQ says no deletion tools have been added for now, so check your client's tool list. Calls are capped per workspace and client per 24 hours by plan unless you add Everything AI, and ClickUp says the limits may change. In Claude Code: claude mcp add --transport http clickup https://mcp.clickup.com/mcp, then /mcp. More in ClickUp vs Flux.

Flux

Flux, task management for AI agent teams, is a hosted kanban board where people and agents work on the same cards. The Flux MCP server runs at https://flux.umin.ai/api/mcp over Streamable HTTP. It is stateless, so every request stands on its own. It authenticates with a Flux API key sent as Authorization: Bearer flux_..., not OAuth, so an unattended agent connects exactly like one in your editor.

It has 12 tools. Six read: list_workspaces, list_boards, get_board, list_cards, get_card, which returns the full card with every comment, and search_cards, which matches card titles and returns up to 20 results. Six write: create_card, update_card (title, description, due date, due-date completion and priority), move_card (within one board), reorder_cards, create_column and add_comment. There is no tool to delete cards, assign people or add labels; the REST API covers more.

Keys always include the Read scope; add Write only for agents that should change the board. A key acts as the person who created it, the write tools need that person to be at least a workspace member, and you can revoke a key at any time. Agents connected by API or MCP never take a seat, and the MCP server is on the Free plan. One limit to know: changes made over MCP are not recorded in the activity log yet. For a deeper look, read how the Flux MCP server works.

How do you connect a server to Claude Code, Cursor or Codex?

Every server in the table is remote, so connecting one means giving your client a URL and a way to sign in.

Choose the endpoint and the sign-in. Take the URL from the table. Use OAuth for an agent you supervise in your editor, and an API key or token for one that runs unattended.
Claude Code. Run claude mcp add --transport http <name> <endpoint>. For an OAuth server, start a session and run /mcp to sign in in the browser. For a key, add --header "Authorization: Bearer <key>". Add --scope project to share the server with your team through a .mcp.json file in the repository.
Cursor. Add an entry to ~/.cursor/mcp.json for every project, or .cursor/mcp.json for one, with a url and, for key-based servers, a headers object. Cursor handles OAuth for servers that need it.
Codex. Add a [mcp_servers.<name>] table to ~/.codex/config.toml with the url. For a key, set bearer_token_env_var to the name of an environment variable that holds it. For OAuth, run codex mcp login <name>.
Test with a read first. Ask the agent to list your projects or boards. When it names them correctly, it is connected, and you can let it write.

For Flux, the Claude Code command is:

claude mcp add flux --transport http https://flux.umin.ai/api/mcp \
  --header "Authorization: Bearer flux_YOUR_API_KEY"

And in Codex, with the key stored in a FLUX_API_KEY environment variable:

[mcp_servers.flux]
url = "https://flux.umin.ai/api/mcp"
bearer_token_env_var = "FLUX_API_KEY"

Keep keys out of files you commit. The Flux MCP page has copy-paste configs for Claude Code, Cline and Cursor and a curl test, and the Claude and Cursor guides walk through each client. When you sign up, Flux creates an API key called Starter Agent, with read and write access (make a Read-only key for reporting agents), and your dashboard shows a "Connect your first agent" panel with the config filled in.

When does a separate shared board make sense?

If your team already plans in Jira or Linear and your agents only touch that tracker, use its MCP server. One source of truth beats two. A separate board earns its place when:

Your team does not live in Jira or Linear and wants one simple board that people and agents share.
Agents run unattended, in CI or on a server, and you want API-key sign-in without asking an admin to enable tokens or moving to a higher plan.
Agents come and go. You want to add agents without paying for a seat each time.
People review agent work. A workflow we recommend: agents move cards to a review column, and a person moves each card to Done. Flux does not enforce this.

That is the job Flux is built for. See how to set up a kanban board for AI agents, with a human review column and one key per agent.

Sources

All pages opened on 18 October 2026.

MCP specification: transports, transports (2025-11-25), authorization, overview
Atlassian: getting started, API token authentication, admin permissions, GA announcement
Linear: MCP docs, launch changelog
GitHub: github-mcp-server, remote server docs, setup guide, GA changelog
Asana: using Asana's MCP server
Notion: Notion MCP, getting started
Trello: connect Trello to AI assistants
monday.com: Platform MCP, integrate with the monday MCP server, compatible clients, mondaycom/mcp
ClickUp: MCP setup, ClickUp's MCP server (limits and FAQ), supported tools
Clients: Claude Code, Cursor, Codex

Project management MCP servers compared.