A task manager for Codex agents

How does Codex work with a task manager?

Codex connects to Kelvia through MCP. Then it can read the issue before touching code, post a short update when it finds the bug, move the task and record the time. That is the whole trick. Useful plumbing, not sentient Jira.

OpenAI's MCP documentation says the Codex app, CLI and IDE extension share configuration on the same host. Set it up once and stop doing the same setup dance in three places.

How do you connect Codex to Kelvia?

Add the hosted endpoint, complete OAuth and check the connection:

codex mcp add kelvia --url https://mcp.kelvia.app/mcp
codex mcp login kelvia
codex mcp list

In an interactive session, use /mcp to inspect the server. Then start with a read:

List my Kelvia boards and the open tasks assigned to me.
Do not make changes.

The equivalent configuration is short enough to audit:

[mcp_servers.kelvia]
url = "https://mcp.kelvia.app/mcp"
auth = "oauth"

Why keep Codex task state outside the chat?

A long agent transcript is a poor project record. It mixes reasoning, command output, abandoned approaches and the final result. The next person has to reconstruct what matters.

A task has a narrower job: title, acceptance criteria, status, owner, discussion and history. Codex can use that state as the contract for the work and leave a concise update when it stops. A later session can read the same task without importing the entire previous conversation.

What task actions can Codex perform?

Kelvia groups its task actions so Codex can load only what the workflow needs:

• Tasks: list, create, update and reorder tasks; read activity and maintain task summaries.
• Collaboration: read and write comments and worklogs, with changes attributed to the calling identity.
• Planning: manage stages, backlog work, due dates, priorities, tags and a personal time-blocked day plan.
• Access: inspect members, manage invitations and apply the same board roles used for people.

For a coding session, the useful minimal toolsets are usually tasks and comments. Loading fewer schemas saves context and reduces the chance of selecting an unrelated tool.

Should Codex use OAuth or a dedicated agent?

OAuth is the quick start for an interactive session you supervise. A dedicated Kelvia agent key is better for recurring or background work. It carries its own identity, belongs only to selected boards and combines read/create/edit/delete scopes with the assigned board role.

Codex stores the environment variable name, not the key itself:

export KELVIA_API_TOKEN='klv_your_agent_key'
codex mcp add kelvia --url https://mcp.kelvia.app/mcp \
  --bearer-token-env-var KELVIA_API_TOKEN

What is a good Codex task workflow?

Keep it boring: read, plan, code, test, update the board. If the prompt sounds like a movie trailer about autonomous swarms, delete half of it.

Read Kelvia task #17 and inspect the repository.
Propose a plan before editing. After approval, implement it and run tests.
If tests pass, add one worklog and a comment with the result and files changed.
Move the task to Review, not Done.

This gives you four gates: read first, show the plan, test the code, update the board. The final product decision stays with a person. Revolutionary concept, I know.

See the full Kelvia MCP setup and tool reference, or compare the workflow with Linear MCP and Trello MCP.

FAQ

Can Codex update a task manager?

Yes. Codex supports MCP servers and can call their read and write tools under the configured approval policy.

Do the Codex app, CLI and IDE extension share setup?

Yes, on the same Codex host they share the MCP configuration.

Can Codex connect to Kelvia with OAuth?

Yes. Add the hosted URL and run codex mcp login kelvia.

Can a Codex agent have separate permissions?

Yes. Agent keys have their own identity, board membership, role, expiry and granular scopes.