Claude Code
Add the remote HTTP server with an Authorization header using the Claude Code CLI.
View setup guideLITSOURCE × MCP
Bring research evidence into your AI workflow.
Find papers for a claim. Check the references in a draft. Give your assistant access to LitSource’s evidence tools, without leaving your conversation to run each search.
Remote MCP · Agent Token
TRY ASKING YOUR ASSISTANT
From a claim to sources
Use LitSource to find papers on whether exercise improves sleep in adults. Show the source links and distinguish supporting, conflicting and uncertain evidence.
From a draft to checked references
Use LitSource to check this reference list: [paste references]. Flag metadata mismatches and unresolved records for my review.
Example prompts, not generated results. Review the sources before citing them.
Choose where you work.
These setup guides use a scoped LitSource Agent Token. Each client needs manual configuration.
Codex
Add the server URL to your MCP configuration and load the token from an environment variable.
View setup guideCursor
Add LitSource to your MCP configuration for a project or your user account.
View setup guide- 1
Create an Agent Token
Sign in to LitSource, open Agent Tokens and select only the permissions you need. Include usage:read for the free check. Store the token securely; do not paste it into a chat.
- 2
Configure your client
Follow your client’s guide using the endpoint below. The token connects calls to your LitSource account and credit balance.
- 3
Ask for your balance
Ask your assistant to call litsource_get_usage. A successful balance response checks authentication and the usage:read permission before you run a paid search or verification.
Use litsource_get_usage to show my LitSource credit balance.
One LitSource credit balance.
A LitSource account and sufficient credits are required for paid tools. Your assistant’s subscription is separate from LitSource credits.
| Operation | LitSource credits |
|---|---|
Find evidencelitsource_search_evidence | 1 credit with results No results or a failed search: 0 credits. |
Verify referenceslitsource_verify_references | 3 credits per 10 references Rounded up: 1–10 references cost 3 credits; 11–20 cost 6. Free accounts can check up to 10 references per call; larger lists require Pro or Trial access. |
Read a saved run or your usagelitsource_get_run · litsource_get_usage | Free Read access is limited by your token permissions and account. |
Know what you are sharing.
Your account, your permissions
Use separate tokens for different clients. Choose the scopes they need and revoke access from Agent Tokens when you no longer need it.
Inputs and results are stored
Search text, submitted references, results and usage records are stored in LitSource for saved runs and account usage. Your AI client also receives tool results; its own data policies apply.
Evidence still needs your judgment
A matching paper does not establish that a claim is true. Review the source, study context and uncertainty. An unresolved reference is not proof of fabrication.
Keep the conversation. Bring the sources.
Pick a client, connect your account, and check your first claim or reference list.
Choose your client