API & AI IntegrationMCP Integration
🤖

MCP Integration

Let AI assistants use Datronis via Model Context Protocol

What is MCP?

The Model Context Protocol (MCP) is an open standard by Anthropic that lets AI assistants (Claude, Cursor, Windsurf, etc.) connect to external services and call real tools. Datronis exposes one of the most capable public MCP servers in the job-search space: any MCP-compatible AI can search real jobs, browse news, explore companies, and query live platform stats for free — and, with a personal access token, build or update a full resume from nothing but a conversation, submit job applications, and check application status, all on your behalf and without writing a single line of integration code.

💡

Quickest way to test

Paste the server URL into Claude Desktop (or any MCP client) and immediately ask: "Find remote senior React jobs" or "Show me the latest tech news."

ℹ️

New to MCP?

See the Datronis for AI Agents overview page first for a plain-language pitch, the full tool list, and more example prompts — come back here when you're ready to wire up a client.

Streamable HTTPMCP 2025-03-26 Spec6 publicznych narzędzi, 5 wymaga klucza API11 narzędzi

Server URL

The MCP server is a single HTTP endpoint. All three HTTP methods are used by the protocol:

FieldTypeRequiredDescription
POST /api/mcpHTTPYesInitialize a new session or send messages to an existing one
GET /api/mcpHTTPNoOpen a Server-Sent Events stream for server-initiated messages (requires Mcp-Session-Id header)
DELETE /api/mcpHTTPNoClose and clean up a session (requires Mcp-Session-Id header)
Server URL
https://api.datronis.com/api/mcp

Claude Desktop Setup

Add the following to your Claude Desktop configuration file. On macOS this is at ~/Library/Application Support/Claude/claude_desktop_config.json.

claude_desktop_config.json
{
  "mcpServers": {
    "datronis": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://api.datronis.com/api/mcp"
      ]
    }
  }
}
ℹ️

mcp-remote

The mcp-remote package bridges Claude Desktop (which uses stdio) to any HTTP MCP server. Install it once with: npm install -g mcp-remote

Cursor / Windsurf / Cline Setup

These editors support HTTP MCP servers natively. Add the server URL in your editor's MCP settings panel:

.cursor/mcp.json (or equivalent)
{
  "mcpServers": {
    "datronis": {
      "url": "https://api.datronis.com/api/mcp",
      "type": "http"
    }
  }
}

Available Tools

Pierwsze 6 narzędzi poniżej jest publicznych (nie wymaga klucza API) — AI może je wywoływać automatycznie na podstawie Twojej prośby. Pozostałe 5 wymaga osobistego tokenu dostępu (zobacz następną sekcję), ponieważ działają na Twoim koncie: wysyłanie aplikacji, odczytywanie Twoich CV, sprawdzanie historii aplikacji oraz tworzenie lub poprawianie artykułów informacyjnych.

FieldTypeRequiredDescription
search_jobstool
"q""country""city""remote_type""job_type""job_level""skills""salary_min""page""limit""api_key (optional)"
NoSearch job listings by keyword, location, remote policy, job type, experience level, and skills. Pass api_key to personalize results (e.g. whether you saved a job). Returns paginated results.
get_jobtool
"slug (required)""api_key (optional)"
NoGet the full details of a job by its slug or ID — including description, requirements, salary, and application URL. Pass api_key to personalize results.
get_newstool
"title""category_id""locale""limit""offset"
NoFetch the latest news articles. Filter by title keyword, category ID, or locale (language). Returns title, summary, date, and URL.
get_categoriestool
"locale"
NoList all content categories on the platform. Used to discover category IDs for filtering news.
search_companiestool
"query""country""industry""is_hiring""limit""page"
NoSearch companies by name, industry, or country. Optionally filter to actively-hiring companies.
get_platform_statstool
NoGet platform-wide statistics: total jobs, companies, news articles, and users.
apply_to_jobtool
"api_key (required)""job_id (required)""resume_id (optional)""cover_letter (optional)"
YesSubmit a job application on your behalf. If resume_id is omitted, your default (or most recently updated) resume is used automatically.
list_my_resumestool
"api_key (required)""query""status""locale""page""limit"
YesList your saved resumes (including their full content), so an AI agent can show you which one will be used to apply, let you pick a different one, or read the current content before updating it.
create_resumetool
"api_key (required)""title""locale""is_default""name""headline""summary""location""contact""experiences""education""skills""projects""certificates""customSections"
YesCreate a resume by giving your AI agent its content directly — name, headline, summary, contact info, work experience, education, skills, projects, certificates. No manual form-filling: describe your background in chat and the agent builds the resume for you. Skips layout/styling entirely; Datronis applies sensible defaults, which you can still adjust afterward in the human builder at /app/resume-builder/.
update_resumetool
"api_key (required)""resume_id (required)""title""locale""status""is_default""full_replace""name""headline""summary""location""contact""experiences""education""skills""projects""certificates""customSections"
YesUtwórz szkic i opublikuj artykuł informacyjny. Otwarte dla każdego konta — token służy do ograniczania częstotliwości i śladu audytowego, a nie jako brama uprawnień. Zawsze najpierw tworzony w języku angielskim; automatyczny klasyfikator bezpieczeństwa musi zwrócić „green”, zanim artykuł zostanie opublikowany (zobacz „Bezpieczeństwo treści i polityka publikacji” poniżej). Publikuje wyłącznie po angielsku, chyba że podano target_locales.
get_my_applicationstool
"api_key (required)""page""limit"
YesPopraw i sklasyfikuj ponownie artykuł, który draft_news odrzucił jako „flagged”. Każda próba jest rejestrowana osobno — nigdy jako cicha zmiana statusu. Tylko konto, które pierwotnie przesłało artykuł, może przesłać go ponownie.
draft_newstool
"api_key (required)""title (required)""body (required)""content_format (html | markdown; default html)""category_id (optional)""target_locales (optional)"
YesDraft and publish a news article. Body may be rich HTML with classes and inline CSS, or Markdown rendered server-side. Open to any account — the token is for rate-limiting and an audit trail, not a permission gate. Always drafted in English first; an automated safety classifier must return "green" before the article goes live (see Content Safety & Publishing below). Publishes English-only unless target_locales is given.
resubmit_newstool
"api_key (required)""news_id (required)""title (required)""body (required)""content_format (html | markdown; default html)""target_locales (optional)"
YesRevise and reclassify an article that draft_news rejected as "flagged". Every attempt is separately logged — never a silent status flip. Only the account that originally submitted the article may resubmit it.
get_missing_news_translationstool
"locale (optional)""page (optional)""limit (optional)"
NoList published English articles and which locale codes each one is still missing. Pass locale to focus on a single language. Read-only — never translates anything itself.
submit_news_translationtool
"api_key (required)""news_id (required)""locale (required)""title (required)""body (required)""content_format (html | markdown; default html)"
YesPublish your own translation of an already-published article into a new locale. Unlike target_locales on draft_news (which asks Datronis's own server to translate), the title/body you pass here is published as-is — you do the translation. Rejected if that locale is already covered.
admin_create_newstool
"api_key (required, admin)""title (required)""body (required)""content_format (html | markdown; default html)""locale (required)""category_id (optional)""published (optional)""target_locales (optional)""translate_to_all (optional)"
YesAdmin-only: publish a news article directly into any locale and optionally queue selected or all missing translations. Translation jobs are durable and continue after the MCP call returns.
admin_edit_newstool
"api_key (required, admin)""news_id (required)""title (optional)""body (optional)""content_format (html | markdown; used when body is supplied)""category_id (optional)""published (optional)"
YesAdmin-only: edit any existing article's title, body, category, or published status, regardless of who created it. No ownership restriction and no classifier. Omit a field to leave it unchanged.
admin_translate_newstool
"api_key (required, admin)""news_id (required)""target_locales (optional; omit for all missing)"
YesAdmin-only: queue server-generated translations for selected missing locales, or omit target_locales to fill every missing language. Existing translations are skipped, and retries are idempotent.
admin_get_news_translation_statustool
"api_key (required, admin)""news_id (required)"
YesAdmin-only: check existing and missing locales plus QUEUED, PROCESSING, COMPLETED, or FAILED state for every translation task. complete=true means all 24 languages exist.

Bezpieczeństwo treści i polityka publikacji dla draft_news

Publikowanie wiadomości przez MCP przebiega według stałego, niemożliwego do obejścia procesu: utworzenie szkicu po angielsku, klasyfikacja, a dopiero potem decyzja, czy i w jakich innych językach opublikować.

  • Krok 1 — Szkic angielski — draft_news zawsze najpierw tworzy angielską wersję artykułu, niezależnie od tego, w jakich językach ostatecznie chcesz go opublikować. Zarówno klasyfikacja, jak i wszelka weryfikacja opierają się na tej angielskiej wersji jako źródle prawdy.
  • Krok 2 — Automatyczna klasyfikacja — Klasyfikator bezpieczeństwa AI odczytuje artykuł i zwraca albo „green” (niekontrowersyjny — nauka, technologia, biznes, sport, rozrywka, kultura, zdrowie, styl życia), albo „flagged” (polityka partyjna, wybory, wojna lub konflikt, drastyczna przemoc, mowa nienawiści, dezinformacja lub cokolwiek, co może być postrzegane jako kontrowersyjne lub dzielące przez szeroką publiczność).
  • Jeśli green — artykuł jest natychmiast publikowany po angielsku, a także tłumaczony i publikowany w każdym locale wymienionym w target_locales.
  • Jeśli flagged — artykuł zostaje utworzony, ale pozostaje nieopublikowany. Wywołanie narzędzia zwraca uzasadnienie klasyfikatora. Nic, co można przekazać do draft_news, nie wymusi publikacji z tego miejsca — nie istnieje flaga obejścia.
  • Krok 3 — Przegląd i ponowne przesłanie (opcjonalnie) — wywołaj resubmit_news z tym samym news_id oraz poprawionym tytułem/treścią. Uruchamia to zupełnie nową klasyfikację poprawki i jest rejestrowane jako osobna próba, niezależna od oryginału. Jeśli poprawka jest teraz „green”, zostaje opublikowana; jeśli nadal jest „flagged”, pozostaje nieopublikowana i może zostać przesłana ponownie.
  • Własność — tylko konto, którego api_key pierwotnie wywołało draft_news dla danego artykułu, może wywołać na nim resubmit_news. Inny (nawet ważny) api_key zostanie odrzucony.
  • Języki są opcjonalne — pominięcie target_locales publikuje tylko po angielsku. Podaj żądane 2-literowe kody języków (np. ["de", "ru"]), aby w tym samym wywołaniu przetłumaczyć i opublikować także w tych językach.
  • Flexible authoring — set content_format to markdown for headings, tables, lists, links, images and code blocks, or html for rich HTML, classes and inline CSS. Markdown is converted to stable HTML on the backend before storage, translation and indexing.
ℹ️

Dlaczego nie można tego pominąć

Brama klasyfikacji istnieje po to, aby każdy mógł korzystać z draft_news bez konieczności ręcznego zatwierdzania każdego artykułu przez człowieka — to właśnie sprawia, że pozostawienie otwartego, nieuwierzytelnionego narzędzia do publikowania treści jako publicznego jest bezpieczne.

💡

target_locales vs. submit_news_translation

target_locales on draft_news/resubmit_news asks Datronis's own server to translate the article for you. submit_news_translation is the opposite: you (the calling AI) write the translated title/body yourself and publish it directly — no server-side translation call happens at all. Use get_missing_news_translations first to find articles that genuinely need a given locale.

Authenticated Tools — Personal Access Tokens

apply_to_job, list_my_resumes, create_resume, update_resume, get_my_applications, draft_news, resubmit_news, submit_news_translation, and all admin_* tools act on your Datronis account, so they need to know who you are. Generate a personal access token from your account settings, then send it either as an Authorization: Bearer HTTP header (preferred) or as the api_key argument on the tool call.

⚠️

Prefer the Authorization header over the api_key argument

If your MCP client lets you configure a header, use Authorization: Bearer <token> on every request instead of passing api_key as a tool argument. An argument has to be reproduced verbatim inside the JSON the calling AI generates for the tool call — which means it passes through the model's own context and shows up in conversation history, tool-call logs, and tracing. A header never does. This isn't theoretical: a real key was corrupted in exactly this way once — the model reconstructing the argument spliced an unrelated local file path into the middle of it. If a header isn't an option in your client, the argument still works (checked second, as a fallback), but treat any key that ever passed through it as more exposed and rotate it more readily.

  • Step 1 — Go to /app/api-keys/ while logged in and click "New Key".
  • Step 2 — Copy the generated key (starts with dtk_) — it's shown only once.
  • Step 3 (preferred) — Configure your MCP client to send Authorization: Bearer dtk_... as an HTTP header on every request.
  • Step 3 (fallback) — If your client can't set custom headers, pass it as api_key in your prompt or tool call instead, e.g. "apply to job X using api_key dtk_...".
  • Revoking — Revoke a key anytime from the same page; it stops working immediately, regardless of which method was used to send it.
ℹ️

One key, every capability

The same personal access token also works with the Datronis Converter API (/api/v1/convert) — you don't need a separate key for MCP tool calls.

⚠️

All admin_* tools need an admin account

The four admin tools check the token owner's account role, not just whether the token is valid. A regular account's token is rejected with "Admin access required" even though it works for non-admin tools. There is no separate admin-specific token type — it is the same personal access token, checked against its account role.

ℹ️

ChatGPT production authentication

The api_key argument is useful for direct/developer-mode testing, but a published authenticated ChatGPT plugin should use the MCP OAuth 2.1 authorization flow. The Datronis token must not be pasted into ordinary prompts; configure authentication at the connection level whenever the client supports it.

Resources

The MCP server also exposes a readable resource — the platform overview documentation — which AI assistants can read to understand the platform before calling tools:

Resource URI
platform://docs/overview

How Sessions Work

The server uses the Streamable HTTP transport (MCP spec 2025-03-26):

  • Step 1 — Client sends POST /api/mcp with an initialize message (no session ID header). Every request, including this first one, must send Accept: application/json, text/event-stream — the server rejects anything else with 406 Not Acceptable.
  • Step 2 — Server creates a session and returns the Mcp-Session-Id header.
  • Step 3 — Client sends a notifications/initialized message (with the Mcp-Session-Id header) to confirm the handshake — the spec-correct next step, expected by well-behaved MCP clients even though this server doesn't currently enforce it.
  • Step 4 — All subsequent tool calls include the Mcp-Session-Id header.
  • Step 5 — Client can open a GET /api/mcp SSE stream for server-initiated notifications.
  • Step 6 — Client sends DELETE /api/mcp to close the session when done.
  • Auto-expiry — Sessions auto-expire after 30 minutes of inactivity.
⚠️

Accept header is required

The server rejects any request that doesn't advertise support for both response formats with 406 Not Acceptable: Client must accept both application/json and text/event-stream. Every request below — not just the first — needs Accept: application/json, text/event-stream alongside Content-Type: application/json.

Manual test with curl
# Step 1: Initialize a session
curl -X POST https://api.datronis.com/api/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -D - \
  -d '{"jsonrpc":"2.0","method":"initialize","id":1,"params":{"protocolVersion":"2025-03-26","capabilities":{},"clientInfo":{"name":"curl","version":"1.0"}}}'

# Note the Mcp-Session-Id in the response headers, then:

# Step 2: Confirm initialization (expected: 202 Accepted, empty body)
curl -X POST https://api.datronis.com/api/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "Mcp-Session-Id: <your-session-id>" \
  -d '{"jsonrpc":"2.0","method":"notifications/initialized"}'

# Step 3: List available tools
curl -X POST https://api.datronis.com/api/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "Mcp-Session-Id: <your-session-id>" \
  -d '{"jsonrpc":"2.0","method":"tools/list","id":2}'

# Step 4: Call search_jobs
curl -X POST https://api.datronis.com/api/mcp \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "Mcp-Session-Id: <your-session-id>" \
  -d '{"jsonrpc":"2.0","method":"tools/call","id":3,"params":{"name":"search_jobs","arguments":{"q":"React","remote_type":"REMOTE","limit":5}}}'

Locale Reference

Parametr locale w get_news i get_categories przyjmuje liczbę od 1 do 24. draft_news/resubmit_news używa zamiast tego 2-literowych kodów (np. „de”, „ru”) w target_locales — tych samych kodów, które widnieją w tej tabeli.

FieldTypeRequiredDescription
1English (en)NoDefault
2Persian / Farsi (fa)NoRTL
3Spanish (es)No
4German (de)No
5Chinese (zh)No
6Russian (ru)No
7French (fr)No
8Arabic (ar)NoRTL
9Hebrew (he)NoRTL
10Japanese (ja)No
11Turkish (tr)No
12Italian (it)No
13Korean (ko)No
14Portuguese (pt)No
15Georgian (ka)No
16Armenian (hy)No
17Indonesian (id)No
18Kazakh (kk)No
19Finnish (fi)No
20Norwegian Bokmål (nb)No
21Dutch (nl)No
22Polish (pl)No
23Swedish (sv)No
24Danish (da)No

Example AI Prompts

Once connected, try these natural language prompts in any MCP-aware AI:

  • "Find remote senior React developer jobs in Europe"
  • "What are the latest tech news articles in English?"
  • "Search for fintech companies actively hiring in London"
  • "How many jobs and companies are on the platform?"
  • "Get the job details for slug 'senior-frontend-engineer-acme-corp'"
  • "List all available news categories"
  • "Apply to the senior-frontend-engineer-acme-corp job using my Datronis account (api_key dtk_...)"
  • "List my resumes on Datronis and tell me which one is my default"
  • "Build me a Datronis resume from my background: 4 years as a backend engineer at Acme (2021-present), B.Sc. Computer Science from TU Berlin, strong in Go and PostgreSQL (api_key dtk_...)"
  • „Utwórz szkic artykułu informacyjnego na Datronis o [temat] i opublikuj go po angielsku i niemiecku (api_key dtk_...)”
  • „Moje ostatnie wywołanie draft_news zostało oflagowane — oto poprawiona, neutralna wersja, prześlij ją ponownie pod news_id 123”
  • "Draft a news article on Datronis about [topic] and publish it in English and German (api_key dtk_...)"
  • "My last draft_news call was flagged — here's a revised, neutral version, resubmit it under news_id 123"
  • "Which published articles on Datronis are still missing a Finnish translation?"
  • "Translate news_id 101 into Finnish and publish it using my api_key"

Schema v1 · Updated September 2026