Starlecta Examples MCP Server

starlecta-examples

Open examples for Starlecta, an influencer analytics platform for brands and agencies in Turkey. Starlecta tracks Instagram creators and the brands they work with (Stories, posts and Reels), and explains metric definitions and scope.

Who this is for: developers and analysts at brands and agencies who want to see how Starlecta's public numbers are shaped, read them from code, or connect them to an AI assistant through MCP. It is not the product and holds no customer data; the live tools it documents are Starlecta's public ones.

Türkçe: Starlecta, Türkiye'deki markalar ve ajanslar için bir influencer analitik platformudur; Instagram içerik üreticilerini ve çalıştıkları markaları (Story, gönderi, Reels) izler ve metrik tanımlarını ve kapsamını açıklar. Bu depo; starlecta.com'daki canlı ücretsiz araçları ve ajan anahtarıyla barındırılan MCP sunucusunu anlatır ("Live tools" bölümü), ayrıca açık örnek verisi üzerinde salt okunur yerel bir MCP sunucusu, çevrimdışı örnek iş akışları ve tamamen sentetik örnek çıktılar içerir. Güvenlik bildirimleri: support@starlecta.com.

Live tools on starlecta.com

Besides the offline examples below, Starlecta runs free tools on real data. Everything in this section is hosted by Starlecta; the code in this repository only calls it. Source of truth for access, limits and endpoints: starlecta.com/auth.md. More: docs, page index llms.txt, MCP server card.

Free web tools (in a browser, no sign-up):

The interactive demo is for people in a browser; do not automate it. Agents can call the live tools without a key; your own key is optional (below). The live tools answer from Starlecta's tracked accounts and show dated, rounded figures; they are not a full list of creators in Turkey.

Hosted MCP server: https://starlecta.com/mcp (Streamable HTTP, stateless). Tools: lookup_influencer, compare_influencers, search_influencers, ai_search_influencers, audit_influencer, trend_cloud, brand_of_the_day. Seven tools (server version 1.1.0). A key is optional: without one, calls are tied to the caller's network. Discovery card: server-card.json. Client setup (Cursor, VS Code, Claude Desktop, Claude Code): live/mcp-configs.

Get your own key (optional; an agent can do this itself, once):

curl -X POST https://starlecta.com/api/agent/keys   -H 'content-type: application/json'   -d '{"label":"<agent or company name, up to 80 characters>","email":"<contact e-mail>","use":"<what the agent will do, up to 200 characters>"}'

The key starts with slk_ and is shown once; Starlecta keeps only a hash. Keys per address and per day are limited, so ask for one and reuse it. A person can use the form instead (English, Türkçe). Keep the key secret: never put it in a page, a public repository or a shared prompt. A key keeps your usage under your own limits. Send it as Authorization: Bearer slk_... (optional) on the MCP server and on the REST endpoints; without it the call is tied to your network. Use a key for the purpose you gave; do not share or resell it, copy the catalogue, build lists in bulk or get around the limits with several keys. Revocation: write to support@starlecta.com.

REST endpoints (the bearer header is optional; OpenAPI: public-pages.json):

• Lookup: GET https://starlecta.com/api/agent/tools/lookup?handle=<instagram handle>
• Compare: GET https://starlecta.com/api/agent/tools/compare?a=<handle>&b=<handle>
• Search: GET https://starlecta.com/api/agent/tools/search?category=<code>&tier=<nano|micro|macro|mega>&followers_min=<n>&followers_max=<n>&er_measured=<true|false> (every filter is optional)
• AI search: POST https://starlecta.com/api/agent/tools/ai-search with {"sentence": "<one sentence, 2 to 200 characters>"}. The sentence is read as the same filters as Search (fast mode only).
• Brand of the Day: GET https://starlecta.com/api/agent/tools/brand-of-the-day?day=YYYY-MM-DD (day is optional; the latest day when omitted). No key needed, like Trend Cloud, and the same per-minute window as the other tools. Aggregates only: no creator handles, names, links or per-creator scores. Dated pages: https://starlecta.com/en/brand-of-the-day/YYYY-MM-DD, with a Markdown twin at the same path plus .md.
• Global Account Check: GET https://starlecta.com/api/agent/tools/audit?handle=<instagram handle>. Reads one public Instagram account live (10,000 followers or more) and returns dated, counted figures; it is a reading of public numbers, not a score.

What an answer holds: each handle is open (public follower count and measured engagement rate, dated), tracked (name, tier, category and a rounded follower band, no metric) or not_listed (no match, or an account that cannot be shown). Search returns a count (shown as fewer than 10 below 10), up to two open cards and the number of locked results; it never lists tracked or locked identities. The Starlecta Quality Score is not part of these tools for named creators. Quote a figure together with its date.

Limits per key, or per network without one (values of 1 November 2026; the response that issues your key lists yours):

429 with Retry-After means a limit was reached: wait that many seconds. 401 with {"error":"invalid_key"} means the key you sent is wrong or revoked. 404 means that tool is switched off. The MCP server reports a wrong key differently: HTTP 200 with isError: true and a text block, not 401.

Try it from the command line (live/agent-tools.mjs, Node 24, no dependencies; it sends your own key from STARLECTA_AGENT_KEY when set, and none otherwise, waits for Retry-After once and stops):

node live/agent-tools.mjs trend-cloud
node live/agent-tools.mjs brand-of-the-day            # latest day; or: brand-of-the-day 2026-11-01
export STARLECTA_AGENT_KEY=slk_YOUR_KEY       # optional: your own key and limits
node live/agent-tools.mjs lookup <handle>
node live/agent-tools.mjs compare <handle-a> <handle-b>
node live/agent-tools.mjs search --tier micro --category beauty --er-measured
node live/agent-tools.mjs ai-search "Turkish beauty creators with 10K to 100K followers"

trend-cloud prints, for example (a real excerpt, computed 2026-10-30T19:17Z; counts change daily):

Trend Cloud, last_7_days, computed 2026-10-30T19:17:01.331943Z
Hashtags:
  #29ekimcumhuriyetbayramı  2657 posts
  #29ekim  2126 posts
Platforms:
  Trendyol  65625 shares
  Hepsiburada  60511 shares

brand-of-the-day prints, for example (a real excerpt, day 2026-11-01, captured 2026-11-02; the featured brand or platform changes daily):

Brand of the Day, 2026-11-01: kitapyurdu.com (platform)
kitapyurdu.com grew its creators 11×. It leads the conversation with 61.3%.
  Creators: 121
  Contents: 416
  Reel plays: 3726
Share of conversation:
  kitapyurdu.com  61.3%
  dr.com.tr  20%
  idefix.com  18.7%

The answer also holds the audience reached (gender, age, city, two top interests), movement and loyalty counts and source / source_tr page links. It never names creators.

A worked example with illustrative answers (AI search, two lookups, a comparison and a shortlist note): live/WALKTHROUGH.md. Terms: terms, privacy.

Offline examples: what you can do with them

Also on starlecta.com: tracking competitor Instagram Stories, sample reports, Turkey.

Try it

Node 24. The workflows have no dependencies and no network.

node workflows/compare-two-creators.mjs fixtures/compare-same-tier.json

# Comparison note: example-creator-a and example-creator-b
...
| Quality score (SQS) | 74 (coverage 92%, confidence high) | 66 (coverage 78%, confidence high) |
| Engagement rate, 90 days | 3.1% (38 posts and reels, as of 2026-10-21) | 2.9% (25 posts and reels, as of 2026-10-21) |

## How they stand

- Quality score: example-creator-a is higher than example-creator-b ("similar" means within 5 points).
- Engagement rate: example-creator-a is similar to example-creator-b ("similar" means within 15% of the higher rate).

When one account is not tracked, the note says so instead of inventing a number (fixtures/compare-one-not-tracked.json):

- Quality score: not comparable (one account is not tracked).
- Engagement rate: not comparable (one account is not tracked).

node workflows/competitor-brief.mjs fixtures/brief-input.json

# Competitor brief: Brand A and 3 tracked brands

Instagram, 2026-09-14 to 2026-10-13 (30 days). Synthetic example data.

| Brand | Verified posts | Left out (no evidence) | Creators | Returning | New |
| --- | --- | --- | --- | --- | --- |
| Brand A | 31 | 64 | 19 | 7 | 12 |
...
Brand A ranks 2 of 4 by verified posts and 2 of 4 by creators.

Full outputs are in expected/. Run the tests with npm test; npm run golden rewrites expected/ after an intended change.

Local MCP server (synthetic edition)

cd mcp-server
npm ci
npm test
MCP_ALLOW_SYNTHETIC=1 MCP_DATA_SOURCE=test/fixtures/edition-from-export npm run stdio

The server speaks MCP over stdio (or stateless HTTP with npm start). It advertises four read-only tools: search_creator_examples, get_creator_example, compare_creator_examples, get_instagram_benchmark. A get_creator_example call for sentetik-guzellik-a on the bundled synthetic edition returns:

Sentetik Güzellik A (sentetik-guzellik-a) · Beauty · Micro · 10K–100K · ~48000 followers
Engagement 3.1234; quality score (SQS): in the customer panel

and an unknown name returns Not found. with Missing: not_in_open_pool, never an invented row. Client setup (Claude Code, Claude Desktop, MCP Inspector): mcp-server/examples/client-configs.md. Details: mcp-server/README.md.

Contents

Data and limits

• The fixtures, expected outputs and the bundled MCP edition in this repository are synthetic. The two real, dated excerpts are the Trend Cloud sample (captured from the hosted server on 2026-10-30) and the Brand of the Day sample (day 2026-11-01, captured on 2026-11-02), both in the live-tools section and in live/WALKTHROUGH.md; every other example answer in live/ is marked illustrative. The workflow fixtures use handles like example-creator-a; Instagram does not allow hyphens in handles, so none is a real account. The MCP edition fixture uses names like synthetic_beauty_a and slugs like sentetik-guzellik-a. They are invented and labelled synthetic, but a handle in the underscore form is valid on Instagram and could exist, so do not read it as a statement about any account.
• The local MCP server in mcp-server/ serves a selected pool of open examples (10 in the bundled edition), not every creator in Turkey. It has no default data source: you point it at a public-data folder or URL, and the open edition files are not published. For live data use the hosted tools above, which answer from Starlecta's tracked accounts with the limits stated there.
• Rules the workflows keep: a comparison uses the bands in the response and never names a winner; rates and scores keep their dates, coverage and missing-data reasons, and null is never shown as zero; measured Reel plays and estimated Story views are kept apart and never added; shared creators count accounts that posted for two brands, not shared audience.
• The quality score is no guarantee of sales or reach. The SQS formula and weights are not published here; the MCP server never returns a score for a named creator, and the SQS values in the workflow output above are invented fixture numbers for the placeholder handles example-creator-a and example-creator-b. Definitions: SQS, engagement rate, Story estimate.

Licence and security

Code: MIT, see LICENSE. The licence covers code only; reuse terms for Starlecta data are separate, see DATA-TERMS.md (draft). Report security issues to support@starlecta.com, see SECURITY.md.