raman-mcp-server
by Ramanji025
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| GIT_TOKEN | No | HTTPS git personal access token | |
| LLM_MODEL | No | Chat model | gemma4:26b |
| NEO4J_URI | No | Bolt URI | bolt://localhost:7687 |
| MCP_KB_ENV | No | Environment name | local |
| NEO4J_USER | No | Neo4j username | neo4j |
| QDRANT_URL | No | Qdrant REST endpoint | http://localhost:6333 |
| GIT_USERNAME | No | HTTPS git username | |
| LLM_PROVIDER | No | ollama / openai / azure | ollama |
| OLLAMA_MODEL | No | Override for enrichment model if LLM_MODEL is not used | |
| OTEL_ENABLED | No | OpenTelemetry tool tracing/metrics | false |
| POSTGRES_DSN | No | Postgres URL | local mcpkb |
| RBAC_ENABLED | No | RBAC enabled | false |
| RERANK_MODEL | No | Reranker id | cross-encoder/ms-marco-MiniLM-L-6-v2 |
| SPARSE_MODEL | No | FastEmbed sparse model | Qdrant/bm25 |
| EMBEDDING_DIM | No | Vector size (must match model) | 384 |
| GRAPH_BACKEND | No | Required 'neo4j' | neo4j |
| LANGFUSE_HOST | No | Self-hosted or cloud Langfuse | http://localhost:3000 |
| OTEL_EXPORTER | No | console or otlp | console |
| RALLY_API_KEY | No | Rally API key for story ids | |
| NEO4J_DATABASE | No | Database name | neo4j |
| NEO4J_PASSWORD | No | Neo4j password | neo4j |
| OPENAI_API_KEY | No | Hosted LLM API key | |
| QDRANT_API_KEY | No | Qdrant API key | |
| RALLY_BASE_URL | No | Rally API base URL | Rally v2 URL |
| RERANK_ENABLED | No | Local cross-encoder rerank | false |
| SPARSE_ENABLED | No | Qdrant BM25 sparse vectors | true |
| TAVILY_API_KEY | No | Optional Tavily; else ddgs | |
| EMBEDDING_MODEL | No | Embedding model id | BAAI/bge-small-en-v1.5 |
| OLLAMA_BASE_URL | No | OpenAI-compatible Ollama | http://localhost:11434/v1 |
| OPENAI_BASE_URL | No | Hosted LLM base URL | |
| LANGFUSE_ENABLED | No | LLM-level tracing (prompts/completions/latency) | false |
| LEARNING_BACKEND | No | file (JSON) or neo4j_qdrant (durable graph + payload boosts) | file |
| LEARNING_ENABLED | No | Online aliases + retrieval boosts from usage | true |
| MCP_KB_HTTP_HOST | No | HTTP bind host | 127.0.0.1 |
| MCP_KB_HTTP_PORT | No | HTTP bind port | 8000 |
| MCP_KB_LOG_LEVEL | No | Log level | INFO |
| MCP_KB_TRANSPORT | No | stdio or http | stdio |
| POSTGRES_ENABLED | No | false = JSON metadata fallback | true |
| AUDIT_LOG_ENABLED | No | Audit log enabled | true |
| MCP_KB_REPOS_ROOT | No | Checkout / index root | ./data/repos |
| PROMPT_GUARD_MODE | No | sanitize (neutralize) or drop (exclude the chunk) | sanitize |
| EMBEDDING_PROVIDER | No | fastembed / sentence-transformers / openai | fastembed |
| GIT_DEFAULT_BRANCH | No | Default branch | main |
| NEO4J_AUTH_ENABLED | No | Set false for passwordless local Docker | true |
| RERANK_CANDIDATE_K | No | Rerank candidate pool | 30 |
| WEB_SEARCH_ENABLED | No | Query-time web gap-fill | true |
| LANGFUSE_PUBLIC_KEY | No | Langfuse public key | |
| LANGFUSE_SECRET_KEY | No | Langfuse secret key | |
| LEARNING_MODEL_PATH | No | LightGBM model file location | ./data/models/ranker.txt |
| EMBEDDING_BATCH_SIZE | No | Embed batch | 64 |
| PROMPT_GUARD_ENABLED | No | Scan retrieved context for prompt-injection phrasing before it reaches the LLM | true |
| MCP_KB_REPOS_MANIFEST | No | Repo list | ./config/repos.yaml |
| LEARNING_MODEL_MIN_AUC | No | Guardrail — model must beat this held-out AUC to be saved | 0.55 |
| LEARNING_TRAIN_ENABLED | No | Enable learning_cron.py scheduled retraining | false |
| OTEL_EXPORTER_ENDPOINT | No | OTLP collector endpoint | |
| SENSITIVE_SCAN_ENABLED | No | Redact secrets at index time | true |
| LEARNING_DASHBOARD_PORT | No | Port for learning dashboard | 8060 |
| LEARNING_MODEL_BOOST_CAP | No | Max +/- boost the model can contribute | 0.5 |
| QDRANT_COLLECTION_PREFIX | No | Collection prefix | mcpkb |
| WEB_SEARCH_MIN_CONFIDENCE | No | Web when pack confidence is below this | 0.45 |
| LEARNING_MIN_TRAINING_ROWS | No | Minimum aggregated rows before training is attempted | 30 |
| LEARNING_DASHBOARD_AUTOSTART | No | mcp-kb-server auto-starts cron trainer + dashboard as background threads | false |
| LEARNING_DASHBOARD_REFRESH_S | No | Dashboard auto-refresh interval | 15 |
| LEARNING_TRAIN_INTERVAL_MINUTES | No | Retraining cadence for the cron loop | 1 |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Server capabilities have not been inspected yet.
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
No tools | |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
This server cannot be deployed