Skillberg Console

The Skillberg MCP connector exposes the Skillberg knowledge graph — ~158K skills, ~20.6K occupations, mapped across ESCO / ROME / O*NET / COMPETENT / SSF — directly inside Claude.ai. Three tools are available; all are read-only and idempotent.

Installing the connector

Inside Claude.ai (one-click)

If you're an Anthropic Pro / Team / Enterprise user, open Claude.ai → Settings → Connectors → Browse, find Skillberg, click Add, and authorize with your existing Skillberg account (or create one in the consent flow). The connector is now available in any Claude.ai conversation.

As a custom remote MCP server

Until the directory listing is live, you can add Skillberg as a custom connector:

Create a Skillberg account at console.skillberg.app and mint an API key from /api-keys.
In Claude.ai → Settings → Connectors → Add custom connector, paste https://mcp.skillberg.app as the URL.
Set the authorization header to Bearer <your-api-key>.

Tools

`search_skills`

Semantic search across the Skillberg knowledge graph. Returns skills matching the query, ranked by cosine similarity.

Arguments:

query (string, required) — free-text search query.
limit (int, default 20) — max results, 1-100.
threshold (float, default 0.1) — minimum similarity score, 0-1.

Annotations: readOnlyHint=true, idempotentHint=true, openWorldHint=false.

`get_skill`

Fetch a single skill by its skillberg://skill/{uuid} URI. Returns the full entity: canonical name, multilingual labels, description, mastery levels, and cross-references to ESCO / ROME / O*NET.

Arguments:

uri (string, required) — skillberg://skill/{uuid} URI from a search_skills result.

Annotations: readOnlyHint=true, idempotentHint=true, openWorldHint=false.

`match_cv_to_jobs`

Score how well a Skillberg user's uploaded CV matches a target occupation. Runs the server-side treat → enrich → matchpipeline: extracts the user's skills from their parsed CV, scores each against the job's required-skills profile, and returns the aggregate match.

Pre-requisite: the user must have already uploaded a CV to Skillberg (via POST /users/{userId}/rawDocuments/treat + /enrich). If no enriched document exists, Bummock returns 404 and the tool surfaces a clear error so Claude can tell the user to upload first.

Arguments:

user_id (string, required) — Skillberg user ObjectId (24 hex chars). Visible at /api-keys once signed in.
job_id (string, required) — Skillberg job ObjectId. The target occupation in the knowledge graph.

Annotations: readOnlyHint=true, idempotentHint=true, openWorldHint=false.

Worked examples

Example 1 — find skills related to a job role

Prompt to Claude:

Quelles sont les 5 compétences principales d'un Data Engineer senior ? Utilise Skillberg.

What Claude does: calls search_skills with query="Data Engineer senior", limit=5.

Sample tool response (truncated):

[
  { "uri": "skillberg://skill/8f4c…", "name": "Apache Spark", "score": 0.91 },
  { "uri": "skillberg://skill/2d1a…", "name": "Data warehousing", "score": 0.88 },
  { "uri": "skillberg://skill/c9e7…", "name": "Stream processing (Kafka)", "score": 0.86 },
  { "uri": "skillberg://skill/4fb0…", "name": "Cloud data platforms (AWS / GCP)", "score": 0.85 },
  { "uri": "skillberg://skill/716e…", "name": "Data modelling", "score": 0.84 }
]

Claude then summarizes the result back to the user in natural language, citing each skill with its name and similarity score.

Example 2 — drill into one skill

Prompt to Claude:

Donne-moi la fiche complète de "Apache Spark" en croisant ESCO et O*NET.

What Claude does: calls get_skill with uri="skillberg://skill/8f4c…" from the previous result.

Sample tool response:

{
  "uri": "skillberg://skill/8f4c…",
  "name": "Apache Spark",
  "labels": { "fr": "Apache Spark", "en": "Apache Spark" },
  "description": "Distributed analytics engine for large-scale data processing.",
  "esco_uri": "http://data.europa.eu/esco/skill/abc123…",
  "rome_codes": ["M1805"],
  "onet_codes": ["15-2051.00"],
  "proficiency_levels": ["awareness","working","expert"]
}

Claude can then construct a richer answer that cross-references Apache Spark across the European ESCO taxonomy, the French ROME, and the US O*NET — something that would otherwise require Claude to know all three taxonomies natively.

Example 3 — onboarding a candidate's CV

Prompt:

Voici l'expérience de ce candidat : "5 ans en backend Python, expert PostgreSQL, build pipelines Airflow". Quelles compétences Skillberg correspondent ?

Claude makes one or more search_skills calls and returns a list of canonical Skillberg skills (each with a URI), so the user can plug them straight into a CV / matching / GEPP workflow.

Example 4 — score a Skillberg user's CV against a job

Pre-requisite: upload a CV via console.skillberg.appbeforehand. The user's ObjectId is visible in the URL after sign-in; pick a target job from the knowledge graph (any get_skill-style listing).

Prompt to Claude:

Mon Skillberg user_id est 65f...0a2 et je vise le job_id 64a...e9f. Donne-moi mon score de matching et explique-moi quelles compétences me manquent.

What Claude does: calls match_cv_to_jobs with user_id="65f...0a2", job_id="64a...e9f". The tool runs the full treat → enrich → match pipeline server-side and returns a JobMatch payload with recommended_skills (skills extracted from the CV) and recommended_matching (per-skill score against the target job).

Sample tool response (truncated):

{
  "job_id": "64a...e9f",
  "user_id": "65f...0a2",
  "matches": [
    {
      "recommended_skills": { "skillbergSkills": ["Apache Spark", …] },
      "recommended_matching": { "score": 0.74,
        "matched": ["Apache Spark", "Airflow"],
        "missing": ["Kafka stream processing", "Delta Lake"] }
    }
  ]
}

Claude then narrates the gap and suggests skills to acquire, referencing the same canonical Skillberg URIs the user can drill into via get_skill.

Roadmap: a future match_cv_text_to_job_text tool will let users paste a free-text CV and a job offer in the chat and get the same score back without pre-uploading. It needs new orchestration on Bummock + Gyre and is queued for the next iteration.

Authorization model

The connector implements OAuth 2.1 with PKCE, RFC 7591 dynamic client registration, and short-lived signed access tokens (1 hour) plus refresh tokens (30 days). On first connection a dedicated API key is minted on your account, labelled Claude MCP — YYYY-MM-DD. Revoke it any time from /mcp/connections — Claude.ai will detect the revoked credential on the next call and prompt a reconnect.

Endpoints

GET https://mcp.skillberg.app/.well-known/oauth-authorization-server — RFC 8414 AS metadata
GET https://mcp.skillberg.app/.well-known/oauth-protected-resource — RFC 9728 RS metadata
POST https://mcp.skillberg.app/register — DCR
GET https://mcp.skillberg.app/authorize — PKCE required
POST https://mcp.skillberg.app/token — code + refresh grants
POST https://mcp.skillberg.app/mcp — MCP JSON-RPC over Streamable HTTP (SSE)

Limits & expectations

Free-tier accounts are rate-limited to 60 calls / minute per API key. Paid plans raise this.
The MCP server is hosted in fr-par (Paris). Expect ~50-100 ms p50 latency from European Claude.ai users; cold-start may add ~1.5 s on the first call after a long idle period.
The knowledge graph is updated on a quarterly cadence (ESCO / ROME release schedule). UUIDs are stable across updates.

Privacy

See /privacy for the full data-handling policy. Notably, Skillberg never receives your prompts — Claude.ai only forwards the structured tool arguments (e.g. the query string passed to search_skills).

Support

Email support@skillberg.app for help with the connector. For API or KG-content questions, attach the request ID returned in any error response.

Skillberg MCP Connector

Installing the connector

Inside Claude.ai (one-click)

As a custom remote MCP server

Tools

search_skills

get_skill

match_cv_to_jobs

Worked examples

Example 1 — find skills related to a job role

Example 2 — drill into one skill

Example 3 — onboarding a candidate's CV

Example 4 — score a Skillberg user's CV against a job

Authorization model

Endpoints

Limits & expectations

Privacy

Support

`search_skills`

`get_skill`

`match_cv_to_jobs`