Eli MCP

Connect Eli to your AI stack.

Your assistant sees the same product as the Eli dashboard: website installation, AI Search demand, approved website improvements, measured conversion, revenue evidence and memory. Start with one product state, then preview and approve the exact next action.

Connect Claude Install in Cursor Install in VS Code

Connect another MCP client

Use the remote Streamable HTTP endpoint below. Eli opens its OAuth sign-in and consent screen; no API key needs to be pasted into an MCP client.

https://hireeli.io/api/mcp/mcp

Claude Code

claude mcp add --transport http eli https://hireeli.io/api/mcp/mcp

ChatGPT

In a supported ChatGPT workspace, enable developer mode, choose Create app, enter https://hireeli.io/api/mcp/mcp, scan the tools, complete OAuth, and review which actions the workspace will allow.

Read before you decide

Ask your assistant to run get_eli_state. Eli returns whether the website is connected, what AI Search evidence is available, whether conversion is ready, what has been measured and what to do next. Nothing goes live until you review and approve it. Read tools never change customer data. If the account belongs to several workspaces, pass the exact workspace ID instead of relying on row order.

get_eli_state

Start here. See installation, AI Search coverage, website conversion readiness, revenue evidence, memory and the one next action.

get_machine_capabilities

See every action your assistant can run, the input it needs, and the safety rule that applies.

get_company_intelligence

Read Eli's source-aware understanding of your company, offer, audience, proof, constraints, gaps and conflicting facts.

get_growth_control_tower

See the evidence-backed next growth intervention, why Eli chose it, how to run it, and how the result will be verified.

get_distribution_plan

Connect approved company context and measured opportunities to the next personalized distribution and conversion action.

get_reddit_intelligence

See the exact Reddit threads returned in saved AI answers, why Eli prioritized them and the human-reviewed next step.

get_high_intent_ai_buyers

See privacy-safe buying signals from AI-referred journeys without claiming an identity the evidence does not support.

get_buyer_context

Read the confirmed need, approved proof and allowed next action for one buyer journey.

get_answer_to_pipeline_report

See recommendations, visits, buyer reports, pipeline and CRM-confirmed outcomes as separate evidence tiers.

get_next_action

Answer what to fix first today, why it matters, what Eli will change, and how Eli will verify it.

list_pending_actions

List every open item in the order Eli will handle it.

get_onsite_change

Show the page before and after the change, the sources and the current status.

get_content_draft

Show the full draft, its sources and whether it is ready to publish.

get_correction

Show the wrong statement, the proposed correction and the sources behind it.

list_tracked_prompts

List the buyer questions Eli checks and the latest answers.

get_monitoring_health

Show the last check, the next check, provider failures and why no new work was found.

get_visibility_overview

Read recent brand visibility across ChatGPT, Perplexity, Claude and Gemini.

get_engine_results

Show recent buyer questions, answers and exact brand mentions.

get_competitor_comparison

Show which competitors appear in the same AI answers.

get_ai_traffic / get_revenue_receipts

Show AI visits, buyer reports and confirmed sales as separate results.

get_agent_test_result / get_work_log

Show the latest website test and Eli's recent work.

search / fetch

Search and read Eli's monthly reports and weekly results.

Operate the loop safely from your stack

Start with get_machine_capabilities to see the exact input for every supported action. Every workspace change requires the machine parameter user_confirmed: true literally plus confirmed_preview_hash copied from the mutation-free preview response. A missing, textual, numeric or stale value is rejected and returns the current preview so your assistant can show it before asking you again.

prepare_next_best_conversion_action

Draft only

Prepare the strongest supported next action for a specific buyer journey without activating it.

activate_conversion_experiment

Required

Activate a reviewed conversion path or experiment under the workspace's plan and safety controls.

record_pipeline_outcome

Required

Attach a supported buyer or CRM outcome to the journey without upgrading its evidence tier.

prepare_highest_priority_onsite_change

Draft only

Choose the highest-priority measured page opportunity and prepare its exact before-and-after proposal.

approve_onsite_change

Required

Approve and apply a reviewed change to the connected website.

revert_onsite_change

Required

Restore the saved version from before a website change.

dismiss_onsite_change

Not required

Remove a proposal from the pending queue without changing the website.

approve_and_publish_content

Required

Approve a checked draft and publish it through the connected CMS, or show the remaining step.

dismiss_content_draft

Not required

Dismiss a draft without publishing it.

revert_content_publication

Required

Restore a supported published page from its saved publication receipt.

approve_correction

Required

Publish an approved correction through the connected website path.

dismiss_correction

Not required

Dismiss a correction without publishing it.

publish_layer

Required

Publish the reviewed AI-readable page and ask search tools to check it again.

prepare_onsite_proposal

Draft only

Prepare a new website-change draft without changing the live site.

run_workspace_operation

Required

Manage monitoring, competitors, tracked pages, content drafts, listings, reviews, revenue evidence and workspace settings.

Safety model

Your assistant cannot ship work without permission

Identity and role: OAuth resolves the signed-in user, workspace membership and owner/admin role before a write.
Preview and literal confirmation: shipping is refused until the exact current resource is shown and the request carries literal confirmation.
Limits and history: plan limits are enforced and every website change is recorded.
Revert: applied own-site changes retain their before-state and expose a dedicated confirmed revert.
Hard boundary: Eli never sends pitches, review requests, directory submissions, social posts or other third-party outreach through this connection. It can change the weekly progress email, but it cannot grant itself permission to publish website changes.
Secrets stay out of chat: website credentials, new API keys and billing changes use Eli's secure signed-in API pages, not assistant arguments.

Give your assistant the workflow

The Eli skill teaches an assistant to read the work, show the preview, ask for approval and report what happened. It also explains how to run a weekly review.

Open SKILL.md Installation and API guide

Try the public scan without an account

The keyless endpoint exposes exactly two tools: scan_visibility and get_started. It uses the public scan cache and public rate limits and cannot access a customer workspace.

https://hireeli.io/api/mcp-public/mcp

MCP and API writes are available only when the workspace plan permits the action. Starter includes bounded, confirmed content and owned-site actions. Publishing hosted corrections, hosted layers and third-party work keeps its higher-plan or human-send gate. No Eli account yet? Run the free scan.