Alation Agent SDK with Model Context Protocol (MCP) support for both STDIO and HTTP transport modes. Provides access to Alation's data catalog, lineage, data quality, and AI agent tools.
The Alation MCP server exhibits significant definition quality gaps across multiple dimensions. Of 16 tools, most have descriptions but many lack depth, parameter type definitions are incomplete or missing, and output schemas are not documented. Tool naming is reasonably clear (verb-noun patterns like 'get_', 'check_', 'analyze_'), but descriptions vary widely in quality, some are verbose (alation_context, alation_bulk_retrieval) while others are minimal (check_data_quality, generate_data_product). Critical issues: (1) several tools have empty input schemas (generate_data_product, get_data_dictionary_instructions) with no visible parameter constraints; (2) output schemas are entirely undocumented across all 16 tools, preventing LLMs from knowing what data structure to expect; (3) parameter descriptions are sparse or generic (e.g., 'Data source ID' for ds_id in check_data_quality lacks format/range guidance); (4) error handling guidance is absent, tools return failures but do not guide LLM recovery; (5) dependency relationships between tools are mentioned in long descriptions but not formalized. The verbose descriptions in alation_context and alation_bulk_retrieval (600+ chars) exceed the token-efficient 50 - 200 char baseline and bury key guidance under boilerplate. Tools like sql_query_agent, query_flow_agent, and custom_agent have only one-line descriptions ('Agent for...') that fail to explain when to use them vs alternatives. This server would require substantial refinement before production use, particularly output schema documentation and error-handling clarification.
LOW-LEVEL TOOL: Direct bulk enumeration of catalog objects with filters. `catalog_context_search_agent` handles most catalog questions automatically. Use `bulk_retrieval` when you need direct access to bulk catalog objects. You MUST call `analyze_catalog_question` first to determine workflow. WHEN TO USE: - User explicitly requests "bulk_retrieval" or this specific tool - Following instructions from `analyze_catalog_question` WHAT THIS TOOL DOES: Fetches complete sets of catalog objects without semantic search. Use for structural enumeration, not concept discovery. Supported: table, column, schema, query Not supported: documentation objects USE CASES: ✓ "List ALL tables in finance schema" ✓ "Get all endorsed tables from data source 5" ✓ "Show tables with PII classification" ✗ "Find sales-related tables" → use `get_context_by_id` (concept discovery) ✗ "Tables about customers" → use `get_context_by_id` (semantic search) PARAMETERS: - signature (required, JSON): For complete signature specification, field options, and filter rules, call `get_signature_creation_instructions` first. - chat_id (optional): Chat session identifier
CRITICAL: DO NOT CALL THIS TOOL DIRECTLY LOW-LEVEL TOOL: Semantic search of Alation's data catalog using natural language. You MUST call analyze_catalog_question first to determine workflow. USE THIS DIRECTLY ONLY WHEN: - User explicitly requests "use alation_context" - Following analyze_catalog_question instructions - User provides a pre-built signature ## WHAT THIS TOOL DOES Translates natural language into catalog queries. Returns structured data about tables, columns, documentation, queries, and BI objects. ## PARAMETERS - question (required): Exact user question, unmodified - signature (optional): JSON specification of fields/filters - chat_id (optional): Chat session identifier for context-aware searches For signature structure: call get_signature_creation_instructions() ## USE CASES ✓ "Find sales-related tables" (concept discovery) ✓ "Tables about customer data" (semantic search) ✓ "Documentation on data warehouse" (content search) ✗ "List ALL tables in schema" → use bulk_retrieval (enumeration) ✗ "Get all endorsed tables" → use bulk_retrieval (filter-based list) See analyze_catalog_question for workflow orchestration. See get_signature_creation_instructions for signature details.
No output schemas documented for any of the 16 tools. LLMs cannot know what fields to expect, what references are available for chaining, or how to parse responses. This prevents downstream tool composition and forces agents to infer structure.
Two tools have empty input schemas: generate_data_product and get_data_dictionary_instructions. No parameters are defined or constrained, making it impossible for LLMs to know what data to provide or how the tool behaves.
Inferred effective spec: 2026-07-28+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 52 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 0 | - | v1 |
Analyze a catalog question and determine the appropriate workflow for answering it.
Intelligent agent for searching and exploring the Alation data catalog using natural language.
Check data quality metrics and scores for tables and data sources.
Execute a custom agent configuration with provided payload.
Generate a new data product specification in Alation.
Get detailed catalog context for specific objects by their IDs.
Retrieve definitions of custom fields configured in Alation.
Get instructions for using the data dictionary and accessing documentation.
Retrieve data products from Alation using direct lookup or search. Parameters (provide exactly ONE): product_id (optional): Exact product identifier for fast direct retrieval query (optional): Natural language search query for discovery and exploration IMPORTANT: You must provide either product_id OR query, never both. Usage Examples: get_data_products(product_id="finance:loan_performance_analytics") get_data_products(product_id="sg01") get_data_products(product_id="d9e2be09-9b36-4052-8c22-91d1cc7faa53") get_data_products(query="customer analytics dashboards") get_data_products(query="fraud detection models") Returns: { "instructions": "Context about the results and next steps", "results": list of data products } Response Behavior: Single result: Complete product specification with all metadata Multiple results: Summary format (name, id, description, url)
Retrieve list of configured data sources in Alation.
Retrieve data lineage information from Alation showing upstream and downstream dependencies.
Get instructions for creating signatures used in bulk_retrieval and other tools.
Agent for analyzing query flows and execution patterns in Alation.
Agent for generating and executing SQL queries based on catalog context.
No error handling guidance. Tools are marked READ_ONLY or WRITE but do not document failure modes, recovery paths, or what LLMs should do on errors. Example: get_lineage with 20 parameters provides no indication of which are optional, what validation failures look like, or how to retry.
Three 'agent' tools (catalog_context_search_agent, query_flow_agent, sql_query_agent) have one-line descriptions ('Agent for analyzing...' or 'Agent for searching...') that fail to distinguish them from each other or explain when to call each one. LLMs cannot reason about tool selection with such thin descriptions.
Verbose descriptions bury actionable content. alation_context and alation_bulk_retrieval descriptions exceed 600 characters, well above the 50 - 200 char efficiency baseline. They include extensive boilerplate (USE CASES sections, DO NOT sections) that wastes tokens and makes key guidance hard to extract.
Parameter descriptions are generic or missing format/range constraints. Examples: ds_id in check_data_quality ('Data source ID') lacks numeric range; table_ids in check_data_quality is an array but does not state max length; direction in get_lineage is a string but does not enumerate allowed values (upstream/downstream/both). LLMs cannot self-correct without explicit constraints.
Optional parameters are not clearly marked as nullable. Many parameters have 'nullable: true' in schema but descriptions do not explain default behavior or what happens if omitted. Example: signature in alation_bulk_retrieval is nullable but no guidance on what the tool does if omitted.
Tool relationships and prerequisites are mentioned in descriptions but not formalized. alation_context states 'You MUST call analyze_catalog_question first' and 'call get_signature_creation_instructions()' but these dependencies are not expressed as structured metadata. LLMs may skip the prerequisite and fail.