Repository navigation
MCP Tools Reference
LineageIQ ships three MCP servers built with the MCP Python SDK (mcp[cli]==1.30.0, FastMCP). Together they define 28 distinct tools: 25 on the local server and 27 on each of the remote and HTTP servers. This page lists every tool defined in the code as of v0.2.0.
| Server (MCP name) | File | Transport | Reads from | Credential |
|---|---|---|---|---|
Local (lineageiq) |
backend/scripts/mcp_server.py |
stdio | Local artefacts in index/, analysis/ and frontend/semantic-lineage/data/; the core catalogue tools read the Warehouse when live reads are on |
None for reads; the scan tools take an api_key argument |
Remote (lineageiq-remote) |
backend/scripts/mcp_server_remote.py |
stdio | A running server's REST API (--base-url or FABRIC_LINEAGE_API_BASE_URL, default http://127.0.0.1:8765) |
api_key argument on each tool except get_scan_status
|
HTTP (lineageiq-http) |
backend/scripts/mcp_http_server.py |
Streamable HTTP at https://<your-host>/mcp
|
The REST API of the server that started it |
Authorization: Bearer <api-key> on every request, including initialize
|
With the demo image, use the local server (docker run -i --rm ghcr.io/largolujan/lineageiq:latest python backend/scripts/mcp_server.py); the image starts the web server with --no-mcp-http, so /mcp is off there. Setup for each client: Quick start and the remote MCP setup guide.
| Tool | Parameters | Local | Remote | HTTP | What it returns |
|---|---|---|---|---|---|
list_workspaces |
name_contains="" |
✓ | ✓ | ✓ | Scanned workspaces, optionally filtered by name |
search_items |
query, item_type="", limit=20
|
✓ | ✓ | ✓ | Items matching a partial name, optionally of one type |
get_item_lineage |
qualified_name |
✓ | ✓ | ✓ | One item's metadata plus its direct upstream and downstream dependencies |
get_object_impact |
query, limit=5
|
✓ | ✓ | ✓ | Source objects (tables, views and so on) matching a name, and what depends on them |
get_related_elements |
qualified_name, direction="both", max_hops=2, limit=100
|
✓ | ✓ | ✓ | Items within a bounded number of hops, from a prebuilt relationship index |
get_objects_for_item |
qualified_name |
✓ | ✓ | ✓ | Every source object one item reads or writes |
get_notebook_code |
qualified_name, max_chars=60000
|
✓ | ✓ | ✓ | A notebook's downloaded source plus the read, write and execute operations parsed from it |
get_business_glossary |
none | ✓ | ✓ | The admin-curated business glossary, linking terms to qualified names | |
get_data_products |
none | ✓ | ✓ | ✓ | Admin-curated data products: name, owner, description and linked items |
Items are identified by qualified_name, which has the form Workspace / Item Type / Item Name.
| Tool | Parameters | Local | Remote | HTTP | What it returns |
|---|---|---|---|---|---|
get_workspace_roles |
workspace_name |
✓ | ✓ | ✓ | Who has which role (Admin, Member, Contributor, Viewer) in one workspace |
get_access_audit |
none | ✓ | ✓ | ✓ | Tenant-wide access, including who is Admin in the most workspaces |
get_governance_scan_overview |
none | ✓ | ✓ | ✓ | Sensitivity labels, endorsement, tags and semantic model RLS roles from the Scanner API, with the scan's diagnostics |
get_sensitivity_overview |
sample_limit=15 |
✓ | ✓ | ✓ | Catalogued objects whose column names match a sensitivity category, by category |
get_content_sensitivity_overview |
sample_limit=15 |
✓ | ✓ | ✓ | Columns with a content-verified sensitivity match, by category (profiled Lakehouse tables only) |
get_column_sensitivity |
workspace_name, lakehouse_name, table_name
|
✓ | ✓ | Content-based sensitivity match percentages for one Lakehouse table, never raw values | |
get_privacy_coverage_map |
none | ✓ | ✓ | ✓ | Privacy locator: one row per flagged table with every matched category |
get_privacy_framework_map |
none | ✓ | ✓ | ✓ | The same locator grouped by legal framework, then by category |
| Tool | Parameters | Local | Remote | HTTP | What it returns |
|---|---|---|---|---|---|
get_data_quality_overview |
none | ✓ | ✓ | ✓ | Tenant-wide pass, fail and not-profiled counts for the admin-defined data-quality rules |
get_data_quality_for_table |
workspace_name, lakehouse_name, table_name
|
✓ | ✓ | ✓ | Rule results for one Lakehouse table |
get_column_profile |
workspace_name, lakehouse_name, table_name
|
✓ | ✓ | Column statistics (null %, distinct count, safe min and max) for a profiled Lakehouse table |
For the profiling and content-sensitivity tools, found: false means "not profiled or checked yet", not "does not exist".
| Tool | Parameters | Local | Remote | HTTP | What it returns |
|---|---|---|---|---|---|
get_execution_stats |
query, limit=10
|
✓ | ✓ | ✓ | Run counts, outcomes and min, average, median and max duration per job type for matching items |
get_failure_hotspots |
limit=20 |
✓ | ✓ | ✓ | Items ranked by run-failure rate |
get_schedules |
query="", workspace_name=""
|
✓ | ✓ | ✓ | Configured run schedules, optionally filtered by item name or workspace |
| Tool | Parameters | Local | Remote | HTTP | What it does |
|---|---|---|---|---|---|
get_scan_status |
none | ✓ | ✓ | ✓ | Status of the current or last scan, with its scan_id
|
get_scan_queue |
none | ✓ | Requests waiting for the running scan (the other servers include the queue in get_scan_status) |
||
trigger_scan |
workspace_name (+ api_key on local and remote) |
✓ | ✓ | ✓ | Starts a scan of one workspace and returns at once with a scan_id, or queues it if a scan is running |
trigger_full_scan |
api_key on local and remote |
✓ | ✓ | ✓ | Starts a full tenant scan, same scan_id contract |
stop_scan |
api_key on local and remote |
✓ | ✓ | ✓ | Stops the running scan |
On the remote server every tool except get_scan_status also takes an api_key argument. Scans are fire-and-forget: poll get_scan_status and compare its scan_id with the one you received.
- The scan tools need an admin-issued API key with
fullscope. Aread_onlykey getsinsufficient_scope. The only exception is when an admin has turned onallow_anonymous_scans. - On the remote and HTTP servers, the governance, quality and privacy tools call
/api/admin/...routes that require a valid API key. There is no anonymous fallback for governance data. - On the local server, read tools need no credential: they read files on the machine where the server runs.
Against the demo data:
- "What feeds the RPT_Sales_Dashboard report?" (
search_items,get_item_lineage) - "What depends on the sales_summary table?" (
get_object_impact) - "Which items fail most often, and how long do they run?" (
get_failure_hotspots,get_execution_stats) - "Who has access to the Contoso-Sales_Prod workspace?" (
get_workspace_roles)