MCP Tools Reference¶

The MCP server provides 12 tools for document ingestion, search, AI Q&A and sign-in (14 with MCP_BROWSER_SIGN_IN=true).

Available Tools¶

Tool	What it does
`get_system_status()`	Verify setup and database connections
`ingest_documents(data_source, skip_graph, enabled_graph, enabled_rdf, ...)`	Process documents from any of the 14 data sources
`ingest_text(content, source_name, skip_graph, enabled_graph, enabled_rdf)`	Ingest and analyze specific text content; `skip_graph=True` skips KG extraction, `enabled_graph` / `enabled_rdf` pick the graphs
`search_documents(query, top_k, source_auth, scope, sources)`	Hybrid search — find relevant document excerpts; `source_auth` filters to what that user may read, `sources` limits to chosen sources
`query_documents(query, top_k, source_auth, scope, sources)`	AI-powered Q&A over your document corpus (same filtering arguments)
`list_sources()`	The sources with documents in the stores, for the `sources` argument
`get_auth_requirements()`	Which sign-ins the stored documents need before search / query
`sign_in(source, username, password, url)`	Sign in to `alfresco`, `nuxeo`, `cmis` or `s3` (IAM access key); returns the credential for `source_auth`
`start_sign_in(source)` / `finish_sign_in(handoff_id)`	Only with `MCP_BROWSER_SIGN_IN=true`: browser sign-in for Microsoft 365, Box, Google or Azure Storage
`test_with_sample(skip_graph, enabled_graph, enabled_rdf)`	Quick system verification with built-in sample content; `skip_graph=True` for vector-only
`check_processing_status(id)`	Track long-running async ingestion tasks
`get_python_info()`	Python environment diagnostics
`health_check()`	Verify backend API connectivity

`ingest_documents()` — Arguments¶

Argument	Type	Default	Description
`data_source`	`str`	`"filesystem"`	Source type — see Data Source JSON Config Strings table below
`paths`	`str`	`None`	`filesystem` only — file/directory paths; JSON array `["p1","p2"]`, comma-separated, or single path
`skip_graph`	`bool`	`false`	Skip KG extraction and graph store writes; chunk + embed + vector/search only
`enabled_graph`	`bool`	`None`	Build the property graph (`PG_GRAPH_DB`); `None` = the server's `DEFAULT_ENABLED_GRAPH`
`enabled_rdf`	`bool`	`None`	Build the RDF graph (`RDF_GRAPH_DB`); `None` = the server's `DEFAULT_ENABLED_RDF`
`enable_sync`	`bool`	`false`	Enable automatic change detection and incremental updates for the source
`<source>_config`	`str`	`None`	JSON config string for non-filesystem sources (e.g. `alfresco_config`, `s3_config`) — see table below

Note

filesystem uses the paths argument, not a JSON config string. All other sources pass their connection details as a JSON string in the corresponding <source>_config argument.

Data Source JSON Config Strings¶

`data_source`	Config Argument	JSON Fields
`filesystem`	`paths` (not JSON)	File/directory path(s) — no config string needed
`alfresco`	`alfresco_config`	`{"base_url": "...", "username": "...", "password": "...", "paths": [...], "nodeDetails": {...}}`
`cmis`	`cmis_config`	`{"cmis_url": "...", "username": "...", "password": "...", "paths": [...]}`
`s3`	`s3_config`	`{"bucket": "...", "aws_access_key_id": "...", "aws_secret_access_key": "...", "region": "..."}`
`azure_blob`	`azure_blob_config`	`{"connection_string": "...", "container_name": "..."}`
`gcs`	`gcs_config`	`{"bucket_name": "...", "credentials_path": "..."}`
`onedrive`	`onedrive_config`	`{"client_id": "...", "client_secret": "...", "tenant_id": "..."}`
`google_drive`	`google_drive_config`	`{"credentials_path": "...", "folder_id": "..."}`
`sharepoint`	`sharepoint_config`	`{"client_id": "...", "client_secret": "...", "tenant_id": "...", "site_url": "..."}`
`box`	`box_config`	`{"client_id": "...", "client_secret": "...", "folder_id": "..."}`
`web`	`web_config`	`{"urls": ["https://...", "https://..."]}`
`wikipedia`	`wikipedia_config`	`{"titles": ["Article Title", "..."]}`
`youtube`	`youtube_config`	`{"urls": ["https://youtube.com/watch?v=...", "..."]}`

`skip_graph` — All Ingest Tools¶

skip_graph=True is available on all three ingest tools:

ingest_documents(data_source="filesystem", paths=["/docs"], skip_graph=True)
ingest_text(content="...", source_name="doc.txt", skip_graph=True)
test_with_sample(skip_graph=True)

When set, the document is chunked, embedded, and stored in vector + search indexes but KG extraction and property graph / RDF store writes are skipped. Useful for fast bulk ingest when graph queries are not needed.

To build only one of the graphs, the same three tools take enabled_graph (property graph) and enabled_rdf (RDF graph); one left out follows the server's defaults (DEFAULT_ENABLED_GRAPH / DEFAULT_ENABLED_RDF, both true unless set):

ingest_documents(data_source="filesystem", paths=["/docs"], enabled_rdf=False)   # property graph only
ingest_text(content="...", source_name="doc.txt", enabled_graph=False)           # RDF graph only

Permission Filtering — Three Levels¶

Level	How	Result
1 (default)	`search_documents` / `query_documents` with no credentials	No filtering. Needs the backend to trust the MCP server: the same random value in backend `TRUSTED_CLIENT_KEY` and MCP server `FLEXIBLE_GRAPHRAG_TRUSTED_KEY`
2	the same tools with `source_auth` (e.g. from `sign_in`)	Those sources are filtered to what that user may read
3	`MCP_BROWSER_SIGN_IN=true` on the MCP server	Adds `start_sign_in` / `finish_sign_in` (a link the user opens in a browser); `AUTH_HANDOFF_CONFIRM_CODE=true` on the backend adds a code to compare first

Details: Full MCP README — Permission filtering and the Sign-In and Permission-Aware Search guide.