The AI dashboard is currently in beta and may be unavailable or have limited functionality in certain regions.

Tutorial

Token Saver for Claude and Codex

Connect Elis Token Saver to Claude Code, Codex, or any OpenAI-compatible MCP host and start using right-sized task plans immediately.

Tip:Production Token Saver MCP requires OAuth. The final MCP tool call runs under your linked Elis account, so paid-program access, private chain recall, and billing checks use the right user.
Tip:The backend scales to zero after quiet periods. Open the startup page first if your MCP client reports a timeout: https://tryelisai.com/mcp.
1

Wake the server

Open the always-on startup page before first connection. It wakes the backend and tells you when to retry your MCP client.

https://tryelisai.com/mcp
2

Pick the endpoint

Use the Claude endpoint for Claude Code and the OpenAI endpoint for Codex or ChatGPT-compatible MCP clients.

Claude: https://api.tryelisai.com/api/mcp/claude
Codex/OpenAI-compatible: https://api.tryelisai.com/api/mcp/openai
3

Install in Claude Code

Paste this command and then run a normal prompt with Token Saver enabled:

claude mcp add --transport http elis-token-saver https://api.tryelisai.com/api/mcp/claude

For campaign tracking, append a referral and UTM source to the server URL:

claude mcp add --transport http elis-token-saver "https://api.tryelisai.com/api/mcp/claude?ref=YOURCODE&utm_source=claude"
4

Use the Codex/OpenAI MCP URL

In a Codex or OpenAI-compatible host that accepts remote MCP servers, add this server URL:

https://api.tryelisai.com/api/mcp/openai

Referral links work the same way:

https://api.tryelisai.com/api/mcp/openai?ref=YOURCODE&utm_source=codex
5

Run the lifecycle tools

Start with start_elis_token_saver_plan, execute the returned next_task in the host, optionally submit your own decomposition when requested, call get_next_elis_task with a compact result and usage, then call finish_elis_plan.

1. start_elis_token_saver_plan({ goal, ref?, utm_source? })
2. submit_elis_task_plan({ plan_id, tasks }) // only when start asks
3. get_next_elis_task({ plan_id, previous_result, usage })
4. finish_elis_plan({ plan_id, final_answer, usage })
6

Chain recall: reuse proven plans (and opt out)

For signed-in users, a plan that finishes successfully is stored privately in your org/personal Elis scope as a proven chain, so a similar future goal replays the known-good task sequence instead of re-planning — fewer wasted host tokens. Every finish response discloses this in chain_tracking. Report honest previous_result.status (ok/failed/blocked): failed tasks are retried, then the remaining plan is revised (transition, plan_revision). A successful finish also returns a ~20-token chain_hint you may remember and pass back on a future start for direct recall — never store full chains in host memory.

// opt a plan out of storage/recall
start_elis_token_saver_plan({ goal, constraints: { no_store: true } })
// direct recall of a proven chain on a similar goal
start_elis_token_saver_plan({ goal, chain_hint: "elis_chain:<sig>:<plan_id>" })
7

Link billing for paid overage

OAuth sign-in is required in production. When a plan returns payment_method_required, open Elis Billing, attach a payment method, then retry the same Token Saver request under the linked account.

https://app.tryelisai.com/billing