Skip to content

Build your workflow

MCP commands

The reference for storing, finding, organizing, and maintaining your project memories.

Command reference

#

MCP Commands Reference

Recall exposes 40+ tools that Claude uses behind the scenes. You don't need to memorize these — just ask Claude in plain English and it picks the right tool. This page is a reference for when you want to understand what's happening or ask for something specific.

Everyday Memory Commands

The commands Claude uses most often — storing, finding, and managing your memories.

CommandWhat It DoesYou Might Say
store_memorySaves a new memory"Remember that we use ULIDs for all IDs"
search_memoriesFinds memories by meaning (semantic search)"What did we decide about the database?"
update_memoryEdits an existing memory"Update the auth decision — we switched to JWT"
delete_memoryRemoves a memory"Delete the memory about the old API format"
store_batch_memoriesSaves several memories at once"Remember all these decisions from today"
check_duplicateChecks if content already exists before storing (pre-write dedup)"Check if we already have this stored"
import_conversationsBulk import from Claude Code, ChatGPT, or text transcripts. Auto-classifies content type and importance."Import my old Claude conversations"

Automatic Context Loading

These run in the background to keep Claude aware of your project context. Claude calls them on its own when configured properly.

CommandWhen It RunsWhat It Does
auto_session_startSession startLoads your recent decisions, patterns, directives, pending to-dos, and active workflows
pick_upReturning to workGenerates a structured resumption brief: last session summary, active workflows, pending todos, recent decisions, and unfinished work signals
quick_store_decisionAfter decisionsSaves a decision with its reasoning and alternatives considered
recall_relevant_contextBefore tasksSearches for memories related to what you're about to work on
analyze_and_rememberAfter discussionsExtracts key insights from a conversation and stores them
summarize_sessionEnd of sessionCreates a snapshot of what was accomplished
should_use_rlmLarge contentChecks if content is too big and needs the RLM pipeline

Task Management

Persistent to-do lists that survive across sessions. Pending items are automatically included in session context.

ActionWhat It DoesYou Might Say
todo_list createCreates a new to-do with title, priority, and tags"Add a to-do to refactor the auth middleware"
todo_list listLists all to-dos, optionally by status"Show me my pending to-dos"
todo_list completeMarks a to-do as completed"Mark the auth refactor to-do as done"
todo_list deferDefers a to-do and tracks how many times"Defer the database migration to-do"
todo_list updateUpdates title, priority, status, or tags"Change the auth to-do to urgent priority"

Pending to-dos are automatically injected into auto_session_start context with priority indicators like [!!!] for urgent and [!] for high priority.

Webhook Event Delivery

External services (Sentry, GitHub, Honeybadger, Slack, Linear, PagerDuty, Stripe, Discord) send webhooks to Recall. Events are queued and injected into your Claude sessions via the stop hook.

How it works

  1. Configure your service to POST to your Recall webhook URL (found in Settings → Webhooks)
  2. First event from a new source is auto-quarantined — whitelist it in the dashboard
  3. Route events to a specific workspace so they arrive in the right Claude session
  4. When Claude tries to exit, the stop hook picks up the next event and injects it as a task

Supported sources

Sources are auto-detected via HTTP headers. Sentry uses sentry-hook-resource, GitHub uses x-github-event, etc. You can also override with ?source=custom-name in the URL.

Sentry setup

Create a Sentry Internal Integration (not the legacy WebHooks plugin) with your Recall webhook URL. Then add an alert rule with the "Send a notification (for all legacy integrations)" action. The legacy plugin's "Test Plugin" button may show SIGNATURE_INVALID — this is a Sentry test quirk and does not affect real event delivery.

Event Prompts

Attach custom instructions to webhook sources so Claude understands your codebase when handling events.

What are Event Prompts?

When a webhook event arrives, Claude receives a default instruction like "Investigate this error and propose a fix." Event Prompts let you replace this with guidance specific to your codebase: which files to check, your incident protocol, known failure modes, or service architecture.

Examples

  • Sentry: "Check src/services/notification.service.ts first. FCM token errors mean the app was uninstalled — mark the token as invalid rather than retrying."
  • GitHub PRs: "Run linting on affected packages, verify test coverage is 70%+, check for breaking changes in shared types."
  • PagerDuty: "Follow incident protocol: acknowledge, assess severity, check recent deploys, post to #incidents."

How to configure

Go to Settings → Webhooks and click the document icon next to any source. Enter your custom instructions (up to 2,000 characters). Leave empty to use the default prompt.

Multiple prompts per service

Need different prompts for different projects using the same service? Add ?source=sentry-kiddocare to the webhook URL. Each unique source name gets its own prompt, workspace routing, and event count. Example:

...?token=sk-xxx&source=sentry-kiddocare

...?token=sk-xxx&source=sentry-payments

Best practices

  • Include file paths and module names Claude should check first
  • Document known failure modes specific to your service
  • Reference team protocols (incident response, deployment steps)
  • Keep prompts concise and actionable — under 500 words

Project Organization

Workspaces keep your memories separated by project. These commands control where memories live.

CommandWhat It DoesYou Might Say
set_workspaceSelects which project's memories to use"Set workspace to this project"
get_workspaceShows which workspace is active"Which workspace am I in?"
convert_to_globalMakes a memory available in all projects"Make that preference global across all projects"
convert_to_workspaceRestricts a global memory to one project"Keep that rule only for this project"
organize_sessionGroups related memories under a session name"Group today's work as 'Sprint 5'"
get_time_window_contextFinds memories from a time range"What did we work on last week?"

Multi-Session Workflows

For tasks that span multiple days or sessions. Workflows automatically thread context so Claude picks up where you left off.

CommandWhat It DoesYou Might Say
start_workflowBegins tracking a multi-session task"Start a workflow for the auth system"
pause_workflowSaves progress between sessions"Pause the workflow, I'm done for today"
resume_workflowReloads workflow context in a new session"Resume the auth system workflow"
complete_workflowWraps up and generates a summary"Complete the workflow with a summary"
list_workflowsShows all workflows and their status"List my workflows"
get_active_workflowShows the currently active workflow"Is there an active workflow?"
get_workflow_contextGets all memories tagged to a workflow"Show me everything from the API redesign workflow"

Advanced Features

Power-user features for organizing, connecting, and maintaining your memories.

Relationships & History

Connect related memories together and track how they change over time.

link_memoriesConnect two memories
get_related_memoriesFind connected memories
unlink_memoriesRemove a connection
get_memory_graphVisualize connections
get_memory_historyView past versions
rollback_memoryRestore older version
invalidateMark relationship as ended (temporal validity)

Valid relationship_type values

Use one of these 7 values when linking memories. Note relates_to (not related_to).

relates_toparent_ofchild_ofreferencessupersedesimplementsexample_of

Templates & Categories

Create reusable memory formats and organize memories into categories for easier browsing.

create_templateMake a reusable format
create_from_templateUse a template
list_templatesShow all templates
set_memory_categoryTag with a category
list_categoriesShow categories
get_memories_by_categoryBrowse a category

Maintenance & Cleanup

Keep your memory store tidy as it grows. Consolidation merges similar memories automatically.

auto_consolidateAuto-merge similar
force_consolidateForce a cleanup now
consolidation_statusCheck cleanup stats
find_duplicatesSpot duplicate memories
consolidate_memoriesManually merge specific
export_memoriesExport as JSON
import_memoriesImport from JSON

Large Content Processing (RLM)

When you need to analyze content larger than 50KB (big log files, long documents), the RLM pipeline breaks it into chunks and processes them sequentially. You don't need to call these directly — just ask Claude to analyze the content and it will use the pipeline automatically when needed.

create_execution_contextStore large content
decompose_taskBreak into subtasks
inject_context_snippetGet relevant chunk
update_subtask_resultSave partial result
merge_resultsCombine results
verify_answerCross-check accuracy
get_execution_statusCheck progress

Keep up with what's changing.

Read the changelog