LitSource Agent Platform

MCP connection guide

Connect an agent to https://litsource.net/mcp over Streamable HTTP with a scoped LitSource Agent Token.

Available tools

litsource_search_evidence — 1 credit only with results.
litsource_verify_references — existing Verify formula.
litsource_get_run — free.
litsource_get_usage — free.

Paid tools require a stable request_id. Reuse it only when retrying the identical call.

Codex

Export LITSOURCE_AGENT_TOKEN, add this to ~/.codex/config.toml, then run codex mcp list. Codex CLI, the IDE extension, and the ChatGPT desktop Codex host share this configuration.

[mcp_servers.litsource]
url = "https://litsource.net/mcp"
bearer_token_env_var = "LITSOURCE_AGENT_TOKEN"

Claude Code

Official setup reference

HTTP is the recommended remote transport. Run the command, then use /mcp or claude mcp get litsource to confirm it is connected.

claude mcp add --transport http --scope user \
  --header "Authorization: Bearer $LITSOURCE_AGENT_TOKEN" \
  litsource https://litsource.net/mcp

Cursor

Official setup reference

Save this as ~/.cursor/mcp.json for global use or .cursor/mcp.json for one project. Cursor expands the token from your environment.

{
  "mcpServers": {
    "litsource": {
      "url": "https://litsource.net/mcp",
      "headers": {
        "Authorization": "Bearer ${env:LITSOURCE_AGENT_TOKEN}"
      }
    }
  }
}

ChatGPT

Current ChatGPT Web custom plugins require an MCP-compatible OAuth 2.1 authorization flow for customer-specific data. LitSource Agent v1 uses a static scoped Bearer token, so direct ChatGPT Web plugin connection is not claimed as supported yet.

You can use LitSource today from OpenAI's Responses API by configuring a remote MCP tool with server_url and an Authorization header, or use the REST API. Native ChatGPT Web support is Phase 2 after OAuth discovery, PKCE, consent, and token refresh are implemented.

OpenAI official OAuth requirements

Security checklist

Never commit the lsa_… token; load it from an environment variable.
Create separate tokens per agent and grant only required scopes.
Grant only the required scopes and revoke a token immediately if exposed.
Use a new request_id for a new paid operation.

What you can do

Verify a reference list in Cursor or Codex. Paste your draft references and ask the agent to run litsource_verify_references — it flags fabricated or mismatched citations before they reach your paper.
Fact-check a claim in Claude. Ask it to call litsource_search_evidence on a sentence and get real PubMed / Crossref evidence supporting or challenging it.
Check your spend. Ask litsource_get_usage for your current balance and today's token usage.

Pricing: litsource_search_evidence costs 1 credit only when results are returned; litsource_verify_references costs 3 credits per 10 references; litsource_get_run and litsource_get_usage are free. Free accounts can verify up to 10 references per call. There is no daily credit limit — your balance is the only cap.