MCP Integration

Real-time catalog verification and knowledge search

Camel-Kit uses the Model Context Protocol (MCP) to provide real-time catalog verification and knowledge search without pre-loading documentation into the agent’s context.

Two MCP Servers

Camel MCP — Catalog Verification

On-demand access to Apache Camel component catalogs, EIP patterns, data formats, and languages.

5 tools:

  • camel_catalog_component — component details, URI format, properties
  • camel_catalog_eip — EIP patterns (choice, split, aggregate)
  • camel_catalog_dataformat — data formats (JSON, XML, CSV, Avro)
  • camel_catalog_language — expression languages (Simple, JSONPath, XPath)
  • camel_validate_route — YAML route syntax and structure validation
Knowledge MCP — Semantic Search

Hybrid semantic search over 166,973 indexed documents — Apache Camel documentation, guides, CVE advisories, and release notes.

2 tools:

  • hybrid_search — 20% BM25 + 80% KNN Vector search
  • lookup_component — pure BM25 for exact component name matching

Embedding model: granite-embedding-small-english-r2 (384-dim, Q8 quantized)

Catalog Tools in Action

camel_catalog_component

Fetch component details — URI format, properties, supported options:

{
  "name": "http",
  "syntax": "http:httpUri",
  "title": "HTTP",
  "properties": {
    "bridgeEndpoint": {
      "type": "boolean",
      "defaultValue": "false"
    },
    "httpMethod": {
      "type": "string",
      "enum": ["GET", "POST", "PUT", "DELETE"]
    }
  }
}

Used by skills to verify every component name before it enters a design spec or YAML route.

camel_catalog_eip

Fetch Enterprise Integration Pattern details with YAML examples:

{
  "name": "choice",
  "title": "Choice",
  "description": "Routes messages based on predicates"
}

- choice:
    when:
      - simple: "${header.type} == 'order'"
        steps:
          - to: "direct:processOrder"
    otherwise:
      steps:
        - to: "direct:processOther"
camel_validate_route

Validates YAML route syntax, structure, and component references:

- route:
    id: order-api
    from:
      uri: "platform-http:/orders"
      steps:
        - unmarshal:
            json: {}
        - to:
            uri: "jms:queue:orders"

Checks: valid YAML, correct DSL schema, components exist, required properties present.

hybrid_search

Semantic + keyword search over indexed documentation:

{
  "query": "How do I configure SSL for HTTP endpoints?",
  "topK": 5
}

Algorithm: 20% BM25 (exact keyword matching) + 80% KNN Vector (semantic similarity).

Returns ranked chunks from component docs, guides, and CVE advisories.

lookup_component

Pure BM25 search for exact component names — no semantic embedding.

Why separate? Component names are exact matches. Semantic similarity doesn’t help (activemq might match kafka semantically but they’re different components).

{ "componentName": "kafka" }

/camel-knowledge

The /camel-knowledge skill is a prescriptive Q&A interface over Knowledge MCP:

User: "How do I enable retries on the Kafka component?"

Agent:
1. Calls hybrid_search(query="Kafka retries", topK=5)
2. Receives relevant chunks
3. Synthesizes answer with source citations
Question Routing

The skill routes questions to the appropriate tool:

  • Component config → hybrid_search + camel_catalog_component
  • Error troubleshooting → hybrid_search (includes CVE advisories)
  • Best practices → hybrid_search (includes best practice guides)
Cross-Agent Equalization

Entirely MCP-driven — no agent-specific logic:

User question → hybrid_search() → ranked chunks → answer

Works identically on all 5 agents (Claude, Bob, Gemini, Qwen, OpenCode). Same questions, same results, regardless of which agent.

Next Steps