MCP / ACCOUNT WORKFLOWS BETA

Generate reviews with Codex.

Public evidence needs no sign-in. With an account token, read your runs, prepare selected papers and generate cited reviews under your plan and credit rules.

Connect

Run this command in your terminal, then start a new Codex session.

codex mcp add litsynth --url https://litsynth.ai/api/mcp

Or add this Streamable HTTP URL in your Codex MCP settings: https://litsynth.ai/api/mcp

Authorize review generation

Create a token in Settings → Security and enable generation.
Set LITSYNTH_MCP_TOKEN in the environment that launches Codex and use the configuration below. Keep secrets out of configuration files and Git.
Confirm paper selection in the workspace, then ask Codex to list runs, prepare papers, generate and read the result.

[mcp_servers.litsynth_account]
url = "https://litsynth.ai/api/mcp"
bearer_token_env_var = "LITSYNTH_MCP_TOKEN"
tool_timeout_sec = 850

Quick reviews cost 25 credits; deep reviews cost 100 and require an eligible plan. Read run status before retrying; the same run reuses existing generation locks and settlement records.

litsynth_list_review_runs → litsynth_prepare_review → litsynth_generate_review → litsynth_get_review_run

Try asking Codex

Check the running LitSynth version and the two latest releases.
Explain why the benchmark failed despite 100% protected recall. Include denominators, manual review, and limitations.
Validate the holdout protocol, then clear its enabled population criterion and confirm validation rejects it.

Available tools

litsynth_get_release_info: Running release and deployment commit
litsynth_list_releases: Dated changes and limitations
litsynth_get_screening_benchmark: Safety, workload, protocol and audit metadata
litsynth_list_benchmark_predictions: Paginated historical decisions and rationale
litsynth_read_benchmark_artifact: Bounded raw text and complete-file SHA-256
litsynth_validate_screening_protocol: Dry-run the application’s protocol contract

Read the changelog

Generate reviews with Codex.

Public evidence needs no sign-in. With an account token, read your runs, prepare selected papers and generate cited reviews under your plan and credit rules.

Connect

Run this command in your terminal, then start a new Codex session.

codex mcp add litsynth --url https://litsynth.ai/api/mcp

Or add this Streamable HTTP URL in your Codex MCP settings: https://litsynth.ai/api/mcp

Authorize review generation

Create a token in Settings → Security and enable generation.
Set LITSYNTH_MCP_TOKEN in the environment that launches Codex and use the configuration below. Keep secrets out of configuration files and Git.
Confirm paper selection in the workspace, then ask Codex to list runs, prepare papers, generate and read the result.

[mcp_servers.litsynth_account]
url = "https://litsynth.ai/api/mcp"
bearer_token_env_var = "LITSYNTH_MCP_TOKEN"
tool_timeout_sec = 850

Quick reviews cost 25 credits; deep reviews cost 100 and require an eligible plan. Read run status before retrying; the same run reuses existing generation locks and settlement records.

litsynth_list_review_runs → litsynth_prepare_review → litsynth_generate_review → litsynth_get_review_run

Try asking Codex

Check the running LitSynth version and the two latest releases.
Explain why the benchmark failed despite 100% protected recall. Include denominators, manual review, and limitations.
Validate the holdout protocol, then clear its enabled population criterion and confirm validation rejects it.

Available tools

litsynth_get_release_info: Running release and deployment commit
litsynth_list_releases: Dated changes and limitations
litsynth_get_screening_benchmark: Safety, workload, protocol and audit metadata
litsynth_list_benchmark_predictions: Paginated historical decisions and rationale
litsynth_read_benchmark_artifact: Bounded raw text and complete-file SHA-256
litsynth_validate_screening_protocol: Dry-run the application’s protocol contract

Read the changelog