Odin MCP
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| ODIN_BUCKET | No | S3 bucket used by the optional direct observation-data tools. | odin-level2 |
| ODIN_AWS_PROFILE | No | AWS profile used for optional direct S3/Parquet data access. | odin |
| ODIN_MAX_LIST_RESULTS | No | Maximum file-list request limit. | 1000 |
| ODIN_MAX_ROWS_PER_READ | No | Maximum rows per read, per file. | 1000 |
| ODIN_MAX_FILES_PER_READ | No | Maximum number of files per read. | 5 |
| ODIN_DEFAULT_PREVIEW_ROWS | No | Default number of preview rows. | 100 |
| ODIN_MAX_COLUMNS_PER_READ | No | Maximum explicit column requests per read. | 20 |
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
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": false
} |
| prompts | {
"listChanged": false
} |
| resources | {
"subscribe": false,
"listChanged": false
} |
| experimental | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| list_research_topics_toolA | List offline Odin research guidance, topic URIs and supporting sources. |
| get_research_topic_toolB | Read a documented plot/method topic and citations; does not analyse data. |
| search_research_guidance_toolB | Search local literature guidance and aliases; return up to 8 cited snippets. No AWS credentials, web request or numerical analysis is needed. |
| list_partition_values_toolB | Discover published L2 partition values over anonymous HTTPS. Apply all provided filters. Year/month are storage labels, not row-time filters. A budget error never represents a complete empty selection. |
| list_parquet_files_toolB | List published HTTPS Parquet URLs and provenance with snapshot-bound pagination. All filters apply, including when earlier partition levels are omitted. Follow next_token with identical filters; errors are not partial coverage. |
| read_parquet_preview_toolB | Read actual Parquet schema and bounded rows over anonymous HTTPS. Supply an exact catalog-listed file_url, or storage partition filters. Whole files are transferred within byte budgets. No quality screening or time filtering occurs. Legacy s3_uri returns migration guidance. |
| read_parquet_columns_toolA | Read selected physical columns, preserving nulls, nanoseconds and kernels. Supply an exact catalog-listed file_url, or storage partition filters. Row/column limits do not reduce full-file HTTPS transfer. Nested arrays are subject to the output budget; no automatic scientific filtering is applied. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
| plan_odin_analysis | Plan a scientifically explicit Odin analysis using source-linked guidance. |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
| dataset_schema_resource | Machine-readable JSON schema for Odin dataset partitions and columns. |
| dataset_schema_guide_resource | Human-readable Odin dataset schema guide in Markdown. |
| research_guide_resource | Start here: source-linked Odin L2 plots, methods and scientific paradigms. |
| research_catalog_resource | Versioned index of bundled research topics; no credentials required. |
| research_sources_resource | Bibliography with source URLs, scope and checked date. |
TDQS
Scored across 7 tools
The two read tools (read_parquet_columns_tool vs read_parquet_preview_tool) overlap somewhat, but descriptions clearly separate column projection from schema/bounded-row preview. The two listing tools operate at different levels (partition values vs file URLs), and the research trio (list/get/search) is cleanly differentiated. Boundaries are mostly clear, with only the paired read tools requiring a careful description read.
Every tool follows the same verb_noun_tool snake_case pattern (read_parquet_columns_tool, list_parquet_files_tool, get_research_topic_tool, search_research_guidance_tool). The uniform '_tool' suffix further reinforces predictability. No mixing of conventions.
Seven tools is well-scoped for a read-only catalog/guidance server, with no obvious redundancy. Each tool occupies a distinct slot: discovery, listing, reading, and guidance lookup.
Discovery (partitions, files), column reads, previews, and research guidance are all covered, which is solid for a read-only data-catalog purpose. Minor gaps exist — no dedicated schema/metadata-only tool and no meaningful server-side row filtering or aggregation, forcing agents to transfer whole files.