Tools Reference - Cognee Documentation

Cognee MCP currently exposes 14 tools for memory management, retrieval, and dataset operations. Unless noted otherwise, parameter names and defaults below reflect the current MCP server implementation.

Some MCP parameters use compact transport-friendly encodings:

datasets and session_ids are comma-separated strings, not JSON arrays.
top_k must be between 1 and 100.
cognify.data accepts either a single string or a JSON array of strings for batch ingestion.

With backend access control enabled, dataset names are resolved against datasets owned by the current user. Shared datasets that the user can access but did not create may not be targetable by name through MCP retrieval tools.

Available Tools

v1.0 Memory Tools (Recommended)

Prefer these tools for new integrations. They map to Cognee’s current main operations.

`remember`

Store content as permanent graph memory or session memory in one call. See Remember.

Parameter	Type	Default	Notes
`data`	`str`	required	Text to store.
`dataset_name`	`str`	`main_dataset`	Target dataset for permanent memory.
`session_id`	`str`	`None`	When set, stores in session cache instead of permanent graph memory.
`custom_prompt`	`str`	`None`	Custom extraction prompt for permanent-memory mode.

`recall`

Retrieve memory with auto-routing and session-aware behavior. See Recall.

Parameter	Type	Default	Notes
`query`	`str`	required	Natural-language query.
`search_type`	`str`	`None`	Optional override for auto-routing.
`datasets`	`str`	`None`	Comma-separated dataset names. Name lookup is owner-scoped.
`session_id`	`str`	`None`	Session-first retrieval scope.
`top_k`	`int`	`10`	Must be between `1` and `100`.

recall currently accepts dataset names, not dataset_ids. If Bob is querying Alice’s shared dataset, datasets="shared_dataset" can fail even when Bob has permission to use it. In that case, either omit datasets to search across all accessible datasets or use the Python SDK / REST API where dataset_ids are supported.

`forget`

Delete a dataset or wipe all memory owned by the current user. See Forget.

Parameter	Type	Default	Notes
`dataset`	`str`	`None`	Dataset name to delete.
`everything`	`bool`	`False`	Set to `true` to delete all user-owned memory.

At least one of dataset or everything=true must be provided.

`improve`

Enrich an existing graph and optionally bridge session memory into permanent memory. See Improve.

Parameter	Type	Default	Notes
`dataset_name`	`str`	`main_dataset`	Dataset to enrich.
`session_ids`	`str`	`None`	Comma-separated session IDs to bridge into the permanent graph.

Legacy Memory Tools

Still available for lower-level control or existing integrations.

`cognify`

Transform ingested data into a structured knowledge graph.

Parameter	Type	Default	Notes
`data`	`str`	required	A single string or a JSON array of strings for batch ingestion.
`dataset_name`	`str`	`main_dataset`	Dataset to populate.
`graph_model_file`	`str`	`None`	Path to a custom graph model file.
`graph_model_name`	`str`	`None`	Class name inside `graph_model_file`.
`custom_prompt`	`str`	`None`	Custom extraction prompt.

If graph_model_file is set, graph_model_name should also be provided.

`search`

Retrieve memory with explicit lower-level search control.

Parameter	Type	Default	Notes
`search_query`	`str`	required	Natural-language query.
`search_type`	`str`	required	Explicit lower-level search type such as `GRAPH_COMPLETION`, `RAG_COMPLETION`, `CHUNKS`, `SUMMARIES`, or `CYPHER`.
`top_k`	`int`	`10`	Must be between `1` and `100`.
`datasets`	`str`	`None`	Comma-separated dataset names. Name lookup is owner-scoped.

search can scope by dataset name only. If the dataset was created by another user and shared with you, datasets="shared_dataset" may not work even if you have access. When you need to search one specific shared dataset, use the Python SDK or REST API with dataset_ids instead.

`prune`

Reset the local MCP-managed store for a fresh start.This tool has no parameters.

prune is not available in API mode.

Retrieval Context Helpers

Use these after retrieval when you need a fuller document view or nearby chunk context.

`get_document`

Retrieve a source document and its chunks.

Parameter	Type	Default	Notes
`document_id`	`str`	required	Document ID. A chunk ID is also accepted and resolves to its parent document.
`include_metadata`	`bool`	`True`	Include metadata in the response.
`max_chunks`	`int`	`0`	Use `0` to return all chunks.

`get_chunk_neighbors`

Retrieve nearby chunks from the same document.

Parameter	Type	Default	Notes
`chunk_id`	`str`	required	Target chunk ID.
`neighbor_count`	`int`	`2`	Number of chunks to fetch on each side.
`include_target`	`bool`	`True`	Include the original chunk in the response.
`direction`	`str`	`both`	One of `both`, `forward`, or `backward`.

get_document and get_chunk_neighbors are not available in API mode.

Interaction Capture

`save_interaction`

Store a user-assistant exchange so Cognee can process it into memory and related rules.

Parameter	Type	Default	Notes
`data`	`str`	required	Interaction content to ingest.

Data Management

`list_data`

List datasets and their data items with IDs for deletion operations.

Parameter	Type	Default	Notes
`dataset_id`	`str`	`None`	Optional dataset UUID. When omitted, lists all datasets.

`delete`

Remove a specific data item from a dataset.

Parameter	Type	Default	Notes
`data_id`	`str`	required	Data item UUID.
`dataset_id`	`str`	required	Dataset UUID containing the item.
`mode`	`str`	`soft`	Either `soft` or `hard`.

`delete_dataset`

Delete an entire dataset by name.

Parameter	Type	Default	Notes
`dataset_name`	`str`	required	Dataset name to delete.

`cognify_status`

Check current and recent pipeline runs for a dataset.

Parameter	Type	Default	Notes
`dataset_name`	`str`	`main_dataset`	Dataset to inspect.
`pipelines`	`list[str]`	`["cognify_pipeline"]`	Optional list of pipeline names to restrict the status view.

delete_dataset and cognify_status are not available in API mode. list_data(dataset_id=...) also requires direct mode for detailed per-item listing.

Usage Notes

For new integrations, start with remember to store data and recall to retrieve it.
Use improve when you want to enrich an existing graph or bridge session memory into permanent memory.
Use forget to remove a dataset or wipe all memory owned by the current user.
Reach for cognify, search, and prune only when you need lower-level control or are maintaining an older workflow.
Use get_document or get_chunk_neighbors after a CHUNKS result when the initial match is too narrow.
Use cognify_status to verify indexing before querying a freshly processed dataset.
In shared-dataset setups, prefer the Python SDK or REST API when you need UUID-based dataset scoping for a dataset the current user did not create.

Next Steps

Client Integrations

Learn how to use these tools with your AI development environment

​Available Tools

​Usage Notes

​Next Steps

Client Integrations

Available Tools

Usage Notes

Next Steps