Slackdump

Accessing the Slackdump Data

Flowchart:

Slackdump MCP available?
  yes → USE IT (stop here)
  no  ↓
Source has slackdump.sqlite?
  no  → REFUSE; tell user: slackdump convert -f database <source>
  yes ↓
SQLite MCP available?
  yes → USE IT (stop here)
  no  ↓
sqlite3 CLI available?
  yes → use sqlite3 CLI (slackdump-sqlite3 skill)
  no  → REFUSE to run

Use the "Slackdump" MCP if it is available. It is read-only and supports all type of Slackdump archives.

If you need to identify the source type, use the following guidance:

• YYYYMMDD and HHMMSS represent the 24-hour format time when the slackdump was invoked to generate this archive (it is created as a first step);

• Source types:

  1. A directory with database: identified by slackdump_[YYYYMMDD]_[HHMMSS]/slackdump.sqlite
  2. A standalone database: identified by slackdump.sqlite
  3. A chunk directory: identified by slackdump_[YYYYMMDD]_[HHMMSS]/*.json.gz
  4. A dump directory: identified by slackdump_[YYYYMMDD]_[HHMMSS]/[CHANNEL_ID].json
  5. An export directory: identified by slackdump_[YYYYMMDD]_[HHMMSS]/dms.json
  6. An export or dump zip file: identified by slackdump_[YYYYMMDD]_[HHMMSS].zip. In the ZIP file, the file structure is similar to dump and export directories described above.

• If you need to understand the source structure in details, use "slackdump-source" skill.

IMPORTANT: You will be accessing the data in read-only mode. You must not UPDATE or DELETE or run any DML or DDL statements. Only SELECT and data-dictionary querying.

• Database path might be given by the agent, otherwise look for slackdump.sqlite file. If there are several files like this in the current project, ask user to choose one of them.

• If falling back to sqlite3 use the "slackdump-sqlite3" skill.

What to pay attention to

This section describes some data structure quirks that you need to be aware of in order to understand what you're dealing with.

Threads

• Threads are nested conversations. Here's how a thread looks like:

  +- parent_message (ts=12345, thread_ts=12345, latest_reply=XXXX)
  | +-- parent_message (ts=12345, thread_ts=12345)    - conversations.replies endpoint returns the starter message, so it MIGHT be duplicated in the archive
  | +-- first_thread_message(ts=12346, thread_ts=12345)
  | +-- second_thread_message(ts=12355, thread_ts=12345)
  | +-- <...>
  | +-- n'th_thread_message(ts=XXXX, thread_ts=12345)
  +- next_non-threaded_message_in_the_channel (ts=24921, thread_ts=<empty>)

• Note that the parent_message appears twice. One is returned during the conversations.history, and a duplicate is returned by conversation.replies endpoints. It might be duplicated in the archive.

• In the channel there may be parent messages with deleted threads, latest_reply field will be set to a special zero-value "0000000000.000000", there will be no thread messages for such threads, so no need to call get_thread tool on MCP:

  +- parent_message (ts=23456, thread_ts=23456, latest_reply=0000000000.000000)
  |  // there will be no thread messages
  +- next_message (can be threaded or non threaded)

• IS_PARENT=TRUE already implies the thread has replies — a thread-lead message with latest_reply = "0000000000.000000" is stored with IS_PARENT=FALSE. You do not need to additionally filter by latest_reply when looking for threads that have replies.

• IS_PARENT=FALSE indicates a thread message that is not a parent message of a thread (thread reply message), or a parent message of a deleted thread.

• Messages with subtype thread_broadcast ("also sent to channel") appear in both the channel history (CHUNK.TYPE_ID=0) and the thread (CHUNK.TYPE_ID=1). When counting or listing channel messages, filter them out to avoid double-counting:

  WHERE JSON_EXTRACT(DATA, '$.subtype') IS NOT 'thread_broadcast'

Files

Some messages may have non empty files array. A file has a "mode" field. Known file modes:

• hosted: normal Slack-hosted file, downloadable;
• snippet: code snippet hosted by Slack;
• space: Slack canvas / huddle space document;
• external: not hosted on Slack servers, is_external will be true, not downloadable;
• tombstone: file was deleted, download_url will be empty;
• hidden_by_limit: on free workspaces, Slack hides files uploaded 90+ days ago, download_url will be empty.

Only hosted, snippet, and space files are downloadable. The others (external, tombstone, hidden_by_limit) have no usable download URL.

NOTE: do not try to download, do the following:

• If the directory with slackdump.sqlite has /__uploads subdirectory, find it using the following pattern /__uploads/[FILE_ID]/*.
• Otherwise: files are not downloaded; WARN user that files are not present, and user can use slackdump tools redownload, if files are needed.