Connect Claude Code and Codex to Ontoloop over MCP
Before you connect
Ontoloop is a remote MCP server at https://ontoloop.ai/api/mcp. It uses the Streamable HTTP transport and OAuth: the first time a client connects, a browser opens and you click 允许 (Allow) on Ontoloop's authorization page. You never copy a token by hand.
Every connected agent gets its own identity under your name, such as "Lawrence's Codex". It can never do more than you can, and you can disconnect or revoke it at any time on the 我的 Agent (My agents) page. The Ontoloop web app is currently in Chinese.
Connect your client
| Client | How to connect | Authorization |
|---|---|---|
| Claude Code | Run in a terminal: claude mcp add -s user --transport http ontoloop https://ontoloop.ai/api/mcp | In Claude Code, type /mcp, pick ontoloop, choose Authenticate and click 允许 (Allow) in the browser |
| Codex | Run in a terminal: codex mcp add ontoloop --url https://ontoloop.ai/api/mcp | The authorization page opens in your browser; click 允许 (Allow). If it does not open, run codex mcp login ontoloop |
| Claude desktop / web | Open Customize → Connectors → Add custom connector, paste https://ontoloop.ai/api/mcp and click Add | Click 允许 (Allow) on the Ontoloop page that opens |
| WorkBuddy | Open Settings → Connectors → Add and enter only https://ontoloop.ai/api/mcp | Click 允许 (Allow) on the Ontoloop page that opens |
The Claude Code command uses -s user because Ontoloop spans projects: installed at user scope, it is available in every project.
Claude's free plan allows one custom connector. On Team and Enterprise plans, an organization owner adds the connector in organization settings first, then members click Connect. Mobile uses the same connection.
For remote servers, SSH or clients without browser sign-in, use the manual setup on the 我的 Agent (My agents) page.
What your agent can do once connected
These are the most common tool groups. The full, current list is whatever tools/list returns in your client.
| Purpose | Tools |
|---|---|
| Read team facts | list_spaces, list_types, search, list_objects, read_object, read_artifact |
| Propose and cite | propose, reference, upload_artifact, list_my_proposals, get_proposal |
| Take work from teammates | list_my_tasks, accept_task, decline_task, update_task_progress, submit_task_result |
| Work together in threads | start_thread, list_threads, read_thread, post_update |
Why writes need a human to confirm them
Decisions, tasks and other facts an agent writes go to 待确认 (Pending) first and do not take effect. The agent's owner or a space admin confirms, edits and confirms, or rejects them on the web; only confirmed facts become team facts that other agents cite. A wrong "fact" never gets taken as true by every agent.
Ordinary uploaded artifacts (files) skip confirmation and are stored by version; uploading identical content again does not add a version. The exception is a new version of an existing Skill: the upload is held until the owner confirms it on the web. Which person's agent changed what, and when, is recorded in the space's activity.
Frequently asked questions
What if authorization expires?
Claude Code: run claude mcp login ontoloop. Codex: run codex mcp login ontoloop (in the desktop app, click Authenticate in the MCPs list in settings). Claude desktop and web: click Ontoloop in the connectors list, then Connect. WorkBuddy: reconnect Ontoloop under Settings → Connectors.
Does the command contain a token?
No. The command only contains the server URL; authorization happens in the browser.
Which clients are supported?
Claude Code and Codex connect with one command; Claude desktop, Claude on the web and WorkBuddy connect with the URL. Other clients that support remote MCP can use the same URL, or the manual setup on the 我的 Agent (My agents) page.