Developer access

Connect your AI assistant to CiteLadder

Use your company’s saved search, Site Health, Search Intelligence, and AI visibility evidence to research opportunities, draft content, and build action plans. The connection is read-only.

https://citeladder.com/mcp

Connect

Codex CLI

codex mcp add citeladder --url https://citeladder.com/mcp
codex mcp login citeladder

Your browser opens CiteLadder. Sign in, review the account-wide read grant, then approve or deny it.

Claude Code

claude mcp add --transport http --scope user citeladder https://citeladder.com/mcp

Run /mcp in Claude Code and complete the browser sign-in.

These commands describe the supported Streamable HTTP and OAuth setup. Client availability still depends on the version and policies installed in your environment.

Check the connection

Send this as an ordinary request after connecting:

Check my CiteLadder connection only. Discover the exposed tools and list my projects. For the project I choose, inventory the available saved datasets with their observation dates and limitations. If Search Intelligence is available, use read_search_intelligence to enumerate its saved datasets. Do not start research or draft content yet.

Example requests

Review the saved search performance for this project. Show the actual date window, compare it with the previous compatible period, and cite the query-page evidence behind your recommendations.
Find technical and internal-link opportunities in the latest completed crawl. Separate measured issues from unknown or inapplicable checks.
Review the latest AI visibility audit. Show which answers mention us, which pages were cited, and keep answer co-occurrence separate from inspected publisher-page presence.
Use the saved business context and evidence to draft a brief. Label every unavailable fact and do not imply that the draft was saved or published.

Data and tools

This reference is generated from the registered server metadata for release 1.1.0. Every listed tool is read-only. list_skills returns native format and capability metadata; it does not distribute instruction bodies.

fetchid

Fetch a full account-authorized record by a citeladder:// URI returned by search.

get_project_business_contextproject_id, sections

Read the project profile, active prompt portfolio, Site Health, demand, opportunities, and latest visibility audit from persisted CiteLadder data.

list_projectscursor, limit

List a bounded page of projects visible to the connected CiteLadder account; follow next_cursor to enumerate the rest.

list_skillsNo inputs

Inspect metadata for native content formats and supported read capabilities.

read_ai_referralsend_date, project_id, range, start_date

Read the persisted AI-referral projection for a project: sessions referred by AI answer engines, their share of traffic, and the sources behind them. Pass start_date and end_date (ISO YYYY-MM-DD) for an explicit window.

read_demandproject_id

Read the latest persisted demand snapshot, coverage, and comparison for a project.

read_integration_statusproject_id

Read which providers are connected to a project, which properties are mapped, how far the history import has progressed, and how far its coverage reaches. This is the read that explains why a projection is empty.

read_opportunitiescursor, limit, project_id, status

Read the highest-priority current opportunities and their persisted evidence references.

read_performancecompare, compare_end_date, compare_start_date, end_date, granularity, project_id, range, start_date

Read the persisted Search Console/GA4 performance projection for a project: clicks, impressions, CTR, average position and their series for a range, with an optional comparison window. Ranges are day, week, month, 3_months, 6_months, last_synced, or custom with start_date and end_date (ISO YYYY-MM-DD).

read_performance_tablecompare_snapshot_id, cursor, dimension, end_date, page_size, project_id, range, snapshot_id, sort, start_date

Read one paged breakdown of the persisted performance projection: query, page, country, device, search_appearance, day, bing_query or bing_page. Pass the snapshot_id a performance read returned, or a range (with start_date/end_date when the range is custom) to resolve it.

read_prompt_portfoliocohort, cursor, limit, project_id, prompt_set_id

Enumerate the complete persisted prompt portfolio with stable pagination, cohort, status, and generation provenance.

read_query_evidencecursor, limit, project_id, query, resolution_outcome, site_url_id, window_end, window_start

Read an exact saved query/page/date window. Missing windows remain unavailable and never fall back to another range.

read_search_datasetcursor, dataset_id, direction, limit, project_id, sort

Read bounded rows from one published dataset using its dataset-specific persisted grain. Aggregate backlink datasets are never presented as backlink edges.

read_search_intelligenceproject_id

Enumerate saved published Search Intelligence datasets, exact target/market scope, coverage, and acquisition status without acquiring data.

read_site_healthproject_id

Read the latest persisted Site Health score and coverage projection for a project.

read_site_linkscrawl_id, cursor, limit, project_id, site_url_id

Read page-level internal-link metrics and bounded captured neighbours from one concrete crawl. Aggregate rows are not individual edges.

read_site_pagescrawl_id, cursor, limit, page_kind, project_id, status

Read bounded page facts, final analysis references, issues, coverage, and applicability from one persisted crawl.

read_visibility_auditaudit_id, completed_baseline, project_id

Read the latest persisted AI visibility audit status, summary, and evidence reference.

read_visibility_resultsaudit_id, cohort, cursor, engine, limit, project_id, prompt_id

Read persisted answers, entity observations, citations, and query-fanout evidence for one concrete audit without rerunning a provider.

read_visibility_sourcesaudit_id, cohort, cursor, engine, level, limit, project_id

Read owner-computed source usage and denominators for one audit, keeping citation occurrence separate from inspected publisher-page presence.

searchlimit, project_id, query

Search account-authorized projects, opportunities, and prompts. Returns stable record URIs that can be passed to fetch.

Understand the data

MCP reads saved CiteLadder evidence, not a live search engine. Dataset availability depends on completed runs and connected sources. Reading never starts a crawl, sync, model call, paid acquisition, repair, or publication.

A summary does not imply access to every raw provider record. Separate GSC query and page breakdowns are not a page-linked query report. Referring-domain and destination-page summaries are not individual backlink edges. AI citations, answer-level brand co-occurrence, and inspected publisher-page presence are distinct observations.

Unavailable, partial, failed, unknown, not applicable, and observed zero remain separate states. Exact date-window requests do not silently fall back to another range.

Access, privacy, and revocation

The connection requests citeladder:read. Each protected read rechecks the account’s current workspace memberships, so adding or removing membership changes project access without waiting for token expiry. System workspaces are excluded.

The hosted tools do not return provider credentials. Your AI client receives the records you ask it to read; review that client’s data policies and your company’s rules before using sensitive information.

CiteLadder provides the data; what your AI client, its provider and your users do with it is your responsibility. Once data is delivered to a client, CiteLadder cannot recall it and, to the extent permitted by law, Cube27 is not liable for its use or misuse. See the Terms of Service.

Removing or disconnecting CiteLadder from an MCP client may only remove the local connection. To revoke the server-side OAuth grant and invalidate its access and refresh tokens, use a client action that calls CiteLadder’s OAuth revocation endpoint. Workspace membership removal blocks project reads immediately.

Troubleshooting

Symptom	What to do
Authentication expired or denied	Reconnect from the client and complete the browser approval. A denial consumes only that pending request.
A project is missing	Confirm that the signed-in account still has a read-capable role in the project’s workspace.
Data is missing or stale	Refresh or acquire it in CiteLadder, then retry. MCP reads never trigger that work.
An exact date window is unavailable	Choose a persisted window reported by the dataset inventory; do not substitute another period.
A requested tool is unavailable	Use the runtime catalog and produce a supported partial analysis. Never invent the missing evidence.
The client cannot complete OAuth	Update the client and confirm that its current mode supports remote Streamable HTTP MCP with OAuth.