MCP Tools

Reference for all MCP tools exposed by the coree server.

The coree MCP server exposes the following tools. Tool names in the protocol are prefixed with the server name - e.g. mcp__coree__search in Claude Code.

Search

`search(query)`

Primary entry point. Searches memories and code simultaneously in a single ranked result set. Use this before reading files or starting a task.

search("rate limiting implementation")
search("why did we change the auth middleware")
search("Config::load signature")

Returns a combined list of memory snippets and code sections ranked by relevance.

`search_code(query)`

Searches only the code index (source files and git history). Use when you specifically want code results without memory noise.

`search_memory(query, limit?, detail?)`

Searches only stored memories. Use when you want to recall prior decisions or gotchas without code results.

Memory management

`store_memories(memories)`

Stores one or more memories. Each memory has:

Field	Required	Description
`title`	yes	Short label, used in search results
`content`	yes	Full content of the memory
`type`	yes	`decision`, `gotcha`, `how-it-works`, `user`, `feedback`, `project`, `reference`
`importance`	no	0.0-1.0, defaults to 0.5. Use >= 0.8 for critical items.
`tags`	no	Array of strings for filtering
`topic_key`	no	Stable key for upsert - if a memory with this key exists, it is updated
`facts`	no	Array of short fact strings extracted from the content
`source`	no	Provenance tag, e.g. `"reviewed"`
`pinned`	no	Boolean - pinned memories always appear in session context

`get_memories(ids)`

Fetches full content for one or more memory IDs. Use after search() returns IDs you want to read in full.

`list_memories(type?, tags?, limit?, detail?)`

Lists stored memories, optionally filtered by type or tags.

`delete_memories(ids)`

Permanently removes memories by ID.

`pin_memories(ids, pin)`

Pins or unpins memories. Pinned memories always appear in session context regardless of relevance score.

`list_stale_memories()`

Lists memories that have not been accessed recently. Useful for reviewing the memory store.

`evict_stale_memories()`

Removes stale memories. Use with care - this is permanent.

Session

`session_context()`

Called at session start. Returns:

Pending captures from the previous session (for review and promotion to memories)
Most relevant memories for the current project
Summary of recently active memory topics

This tool is called automatically by the hooks installed with the Claude Code and Gemini plugins.

`capture_note(summary, context?)`

Stores a note for review at the next session start. Use for tentative observations during exploration - things you are not ready to commit as a full memory. Captures are reviewed by session_context() and can be promoted to memories or discarded.

Code intelligence

`get_symbol(name, file_path?)`

Looks up a symbol by name. Returns:

Function/type signature
File path and line range
Git line-range history (which commits touched this code)
Churn count (number of commits)
Hotspot score (recent modification frequency - higher means more volatile)
Cross-type similar results (functions/types with similar names)

file_path is optional but narrows the search when the symbol name is ambiguous.

Remote

`remote_sync()`

Triggers a manual sync between the local replica and the remote Turso database. Only relevant when remote_mode = "replica".

Agent integration

`install(target, scope?)`

Installs coree for a supported agent target. Writes the appropriate config files.

Target	Scope	Writes
`cursor`	`project`	`.cursor/mcp.json`
`cursor`	`user`	`~/.cursor/mcp.json`
`claude`	-	Returns instructions (use marketplace instead)
`gemini`	-	Returns instructions (use extension instead)

Scope is required for cursor.

`uninstall(target, scope?)`

Reverses what install wrote.

Diagnostics

`diagnose()`

Reports the current state of the MCP server without requiring the database to be ready. Use when something is wrong.

Returns:

Server state (Syncing / Ready / Failed)
Database path and backend type
Error from crash.log if available
Pattern-matched failure reason and remediation steps
Memory and index counts (if available)