Anara over MCP
Connect another AI assistant to your Anara workspace.
Anara's remote MCP server lets another AI assistant work with your library and start research using Anara. It is separate from the public docs MCP server and from MCP apps inside Anara.
Connect
Use a client that supports remote Streamable HTTP MCP with OAuth. Add a server with this URL:
https://anara.com/api/mcpComplete the browser sign-in and review the consent screen. In clients that ask for a custom client ID or secret, leave those fields blank unless the client requires its own configuration. The exact place to add a server depends on your AI tool and its plan.
Access and permissions
The connection can only use workspace content your account is authorized to access. It can search and read sources and start Anara research. Read-only access is the default; optional write permission allows supported creation and editing actions. It does not grant unrestricted permission to delete, move, or rename library items.
Review the consent scopes before connecting. Do not paste access tokens into chats or public configuration files.
Available tools
| Tool | Purpose |
|---|---|
anara_list_workspaces | List accessible workspaces and identify the default |
anara_shell | Search, read and work with the library through the anara command line in a sandbox, under the granted permissions |
anara_spawn_agent | Start a research task and return a run identifier |
anara_check_agent | Check research status and incremental progress |
anara_get_agent_result | Read the completed result |
anara_cancel_agent | Request cancellation of a research run, then check status to confirm it stopped |
The old anara_execute and anara_run_code tools are retired; anara_shell replaces them. Reconnect or refresh your client's tool list if it still offers old tool names. A research start returns a run identifier, not necessarily the finished answer; the client needs to check progress and retrieve the result.
A research start can name a model. If your plan cannot use that model, or the model ID is not recognized, the research runs on Auto instead of failing. The result names the model that answered in modelName and explains the difference in modelDowngrade. If the model was replaced mid-run because its provider was failing, the result says so in modelFailover.
Troubleshooting
Confirm the endpoint above and complete OAuth sign-in. If access expires, authenticate again. If the wrong library appears, check the selected workspace. Anara research uses your account's applicable usage rules; see usage and credits.
For public paper search in your own application, see the API reference.