datalens-dev-mcp
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| YC_IAM_TOKEN | No | Your Yandex Cloud IAM token (alternative to DATALENS_IAM_TOKEN). | |
| DATALENS_ORG_ID | No | Your Yandex Cloud organization ID. | |
| DATALENS_ENV_FILE | No | Path to environment file containing credentials. | |
| DATALENS_IAM_TOKEN | No | Your IAM token (alternative to YC_IAM_TOKEN). | |
| DATALENS_API_VERSION | No | API version. | auto |
| DATALENS_API_BASE_URL | No | Base URL for DataLens API. | https://api.datalens.tech |
| DATALENS_MCP_LOCAL_CONFIG | No | Path to local configuration file. | |
| DATALENS_MCP_ENABLE_WRITES | No | Enable write operations. | 0 |
| DATALENS_MCP_LIVE_ALLOW_SAVE | No | Allow save operations. | 0 |
| DATALENS_MCP_ENABLE_EXPERT_RPC | No | Enable expert RPC. | 0 |
| DATALENS_MCP_LIVE_ALLOW_PUBLISH | No | Allow publish operations. | 0 |
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
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| dl_server_infoA | Return active process/package identity, capability revision and installed version comparison without accessing DataLens or the project. |
| dl_auth_checkA | Run a harmless API probe and return classified secret-free authentication or scope errors; reuse a failed automatic refresh until explicit recovery. |
| dl_auth_refreshA | Refresh IAM once using the existing yc account/profile. By default yc may open the system external browser for its configured SSO sign-in (up to 120 seconds); cached sign-in can finish automatically. Update API/SDK credentials and verify API access without returning tokens or login URLs. Set allow_browser=false only when browser sign-in is disallowed (15-second noninteractive attempt). |
| dl_method_schemaC | Return the versioned backend and effect contract for one supported DataLens operation. |
| dl_workbooks_listB | List DataLens workbooks with bounded pagination and an explicit completeness marker. |
| dl_workbook_entriesB | Discover workbook entries or establish inventory completeness across numeric pages (up to 200 per page). Retains scope, renderer subtype, typed identity and completeness. A known target uses dl_object_get directly; a partial listing cannot prove absence. |
| dl_object_getA | Read a known target by ID, branch and optional revision. chart (alias widget) accepts confirmed Editor/Wizard/QL subtypes and returns observed_identity; explicit subtypes remain strict. Mutations require the exact observed type. Use summary/projection for inspection, full state for preservation. |
| dl_object_revisionsA | Read one bounded provider-ordered revision page for a known entry, optionally restricted to known revision IDs. Use the returned opaque continuation unchanged. Current saved/published flags are not a multi-page snapshot; absence in limited history cannot prove a write did not apply. |
| dl_object_relationsA | Read direct API-visible relations: direction=from for dependencies, to (legacy default) for consumers. Completeness covers the selected pagination, not a transitive or revision-bound graph. Optional publish_targets with intended saved revisions produces a read-only selected-dependency manifest; no automatic publication. |
| dl_dashboard_snapshotC | Read one dashboard and only its direct dependencies; keep an optional style reference separate. |
| dl_dataset_validateA | Validate Dataset fields locally, or set provider=true to validate a saved Dataset through the provider without saving. Optional exact refresh_source_ids refresh source schemas without forcing field replacement. include_provider_state returns the full validation candidate for an explicit revision-guarded save; this is not a saved result or chart execution. |
| dl_dataset_previewC | Run a bounded Dataset query by exact field GUIDs; this is data evidence, not chart branch proof. |
| dl_authoring_defaultsC | Resolve effective presentation for one family. For a registered recipe, recipe_contract exposes the compiler's bindings_schema, supported source routes, technology and distinct create/static/compiled-update inputs. |
| dl_compile_recipeA | Compile a typed recipe to a compact local artifact reference; never access or write DataLens. First read dl_authoring_defaults(family=recipe_id).recipe_contract for bindings_schema, source routes and distinct create/static/compiled-update inputs. |
| dl_editor_validateA | Validate one Editor draft or a mixed typed draft batch, including artifact references, without execution. With destination and output_dir, prepare exact typed-dashboard wire in a private local artifact (static_validity_only, zero provider calls). An Editor draft uses the observed variant and filename-keyed tabs (meta.json, params.js, etc.), not raw provider data keys. Required tabs depend on subtype; a static draft does not replace the full saved snapshot. |
| dl_object_diffB | Read one saved object and return a compact diff without mutation. Supply patch or, for native dashboard tabs, dashboard_patch with addressed items/layout/connections. Set include_proposed=true only when the full merged snapshot is needed. |
| dl_object_createB | Create typed DataLens drafts or local JSON artifact references ({artifact_path: absolute path}) in dependency order, save, and read back. A typed Dataset uses top-level object_type/name/client_ref and nested dataset.connection_id/source/fields. References may override client_ref/depends_on only. |
| dl_object_updateA | Apply narrow saved-object patches with revision checks and readback; exact no-ops do not write. For existing dashboard tabs use dashboard_patch plus expected_revision: items/globalItems keyed by id with optional namespace, group_control children via group keyed by native id, layout by i, connections by from/to, with add/update/remove and small tab metadata patches. Send only changed records, not whole tabs. Named global parameter removal uses remove_global_params. Compiled Editor updates use artifact_path. Writes retain private receipts; unknown effects are not replayed. |
| dl_object_publishB | Publish chart or dashboard objects only from a fresh verified saved revision, then read the published branch. |
| dl_operation_getB | Read one compact modifying-operation record by exact ID. |
| dl_operation_reconcileB | Reconcile by exact target readback without replay. ACKed historical updates may resolve with unverified content; new changes require fresh CAS. Unknown effects and connection provisioning remain held. New workbook dashboard creates recover by a persisted metadata marker and full readback, without replay. investigate_create=true also investigates legacy unmarked creates by name; names or absence never establish effect. |
| dl_backup_exportC | Export exact object snapshots and a completeness manifest to local files; this is not a full-restore claim. |
| dl_cleanup_previewB | Compute dependency-based preserve and delete sets without deleting anything. |
| dl_cleanup_applyB | Delete exact ordered objects from an unchanged preview. repeat_of asserts separate informed user authorization for one new attempt after an unknown delete; it never enables automatic retries. |
| dl_admin_inventoryA | Read documented license inventory and limits; report that license revoke is unsupported. |
| dl_admin_assign_licensesB | Apply explicit documented license assignments; this operation does not support revoke. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 26 tools
Most tools have clearly distinct purposes, but a few read/validation tools (dl_object_get, dl_dashboard_snapshot, dl_object_relations; dl_editor_validate, dl_dataset_validate) target overlapping resources or operations and could be confused. The detailed descriptions help, but subtle boundaries remain.
All tools use a consistent dl_ prefix and snake_case with a predictable domain_action pattern. Minor variations like dl_workbooks_list versus dl_object_get do not break the overall convention.
26 tools is heavy for an MCP server, and the granularity (separate auth check/refresh, cleanup preview/apply, operation get/reconcile) adds cognitive load. Each tool serves a distinct step, but the set is borderline over-scoped.
The surface covers core DataLens development workflows: auth, discovery, CRUD, validation, publishing, cleanup, backup, and admin. Minor gaps include no explicit generic delete (cleanup_apply requires a preview) and no chart execution/preview tool.