Skip to main content

Summary

The Conversation Archives MCP server provides read-only access to archived message summaries for a conversation. Archives are automatically created periodically to preserve conversation context while managing memory. Requires conversation API key for authentication.

Authentication

This server requires a conversation API key. The key is automatically provided when tools are called within an Agentic Loop.
# Requires conversation API key
curl https://app.chatbotplatform.io/mcp/archives/call/list-archives \
  -H "Authorization: Bearer conv_xxxxx..."
Automatic in Agentic Loops: When using this server within an Agentic Loop, the conversation API key is provided automatically - no manual setup required.

Use Cases

Recalling Past Conversations

User: “What did we talk about last week?” Tool Calls:
  1. Time: relative-time with expression “last week” → Returns date range
  2. list-archives with start_date and end_date from time tool
  3. get-archive for specific archives to show full summaries
Result: Summaries of conversations from last week

Finding Specific Information

User: “Didn’t you recommend a restaurant in our conversation a few days ago?” Tool Calls:
  1. Time: relative-time with expression “3 days ago” → Returns specific date
  2. get-archive with date parameter
  3. Parse archive for restaurant recommendation
Result: Restaurant information from the archived conversation

Browsing Conversation History

User: “Show me a summary of our recent conversations” Tool Calls:
  1. list-archives with limit 10 → Shows last 10 archive summaries
Result: Table of recent archives with dates, message counts, and summary previews

Available Tools

list-archives

List archived message summaries for the authenticated conversation.
ParameterTypeRequiredDescription
start_datestringNoFilter archives from this date (YYYY-MM-DD format)
end_datestringNoFilter archives until this date (YYYY-MM-DD format)
limitintegerNoMaximum number of archives to return (1-100, default: 30)
Returns: Archives ordered by date (newest first) with date, message count, and summary preview for each. Example - All Recent Archives:
Input: limit: 10

Output:
  Recent Archives (10 results):

  | Date | Messages | Summary |
  |------|----------|---------|
  | 2026-02-03 | 15 | User asked about Python data analysis libraries. Discussed pandas, NumPy... |
  | 2026-02-02 | 8 | User requested help with SQL query optimization. Covered indexing... |
  | 2026-02-01 | 12 | Conversation about React component patterns. Discussed hooks, context... |
Example - Date Range Filter:
Input:
  start_date: "2026-01-15"
  end_date: "2026-01-31"
  limit: 50

Output:
  Archives from 2026-01-15 to 2026-01-31 (12 results):

  | Date | Messages | Summary |
  |------|----------|---------|
  | 2026-01-31 | 20 | Detailed discussion about authentication patterns in web apps... |
  | 2026-01-29 | 10 | User asked about Docker deployment best practices... |
  | 2026-01-27 | 15 | Conversation focused on API rate limiting strategies... |
  ...
Combine with Time Utilities: Use the Time MCP server’s relative-time or today tools to convert natural language dates (“last week”, “last month”) to YYYY-MM-DD format for filtering.

get-archive

Retrieve the full details of a specific archive by ID or date.
ParameterTypeRequiredDescription
archive_idintegerNo*The specific archive ID to retrieve
datestringNo*Fetch the archive for a specific date (YYYY-MM-DD format)
Returns: Complete summary text, message count, and metadata.
Either/Or Requirement: You must provide either archive_id or date, but not both.
Example - By Date:
Input: date: "2026-02-03"

Output:
  Archive for 2026-02-03:

  ID: 15789
  Date: 2026-02-03
  Messages: 15
  Created: 2026-02-04 00:05:00 UTC

  Summary:
  The user asked for recommendations on Python libraries for data analysis.
  We discussed several popular libraries:

  1. pandas - for data manipulation and analysis with DataFrames
  2. NumPy - for numerical computing and array operations
  3. Matplotlib - for data visualization
  4. Seaborn - for statistical visualizations
  5. Plotly - for interactive charts

  The user was particularly interested in pandas for working with CSV files.
  I provided code examples for reading CSVs, filtering data, and basic operations.

  The user also asked about performance considerations when working with large
  datasets. I recommended using chunking for very large files and suggested
  considering Dask for datasets that don't fit in memory.

  Metadata: {}
Example - By Archive ID:
Input: archive_id: 15789

Output:
  (Same output as above)
Archive Not Found: If no archive exists for the specified ID or date, an error is returned.

Common Workflows

Recalling Recent Context

1

Get Current Date

Use Time: today to establish current date context
2

Parse User's Time Reference

Use Time: relative-time to convert “last week”, “3 days ago”, etc. to dates
3

List Archives

Use list-archives with date range from time tools
4

Retrieve Specific Archive

Use get-archive for detailed summaries

Searching for Specific Information

1

Determine Timeframe

Ask user when the conversation happened or estimate based on context
2

List Archives in Timeframe

Use list-archives with appropriate date range
3

Review Summaries

Check summary previews to identify relevant archives
4

Get Full Archive

Use get-archive to retrieve complete summary text
5

Extract Information

Parse the archive summary for the requested information

Browsing History

1

Show Recent Archives

Use list-archives with no filters to show recent conversation history
2

User Selects Archive

User identifies which archive they want to explore
3

Show Full Summary

Use get-archive with the selected date or ID

Archive Generation

Archives are automatically created by the platform to preserve conversation context:

When Archives Are Created

  • Periodically: Archives are generated at regular intervals (typically daily)
  • Context Management: When conversation history exceeds configured limits
  • Manual Archival: System administrators can trigger archive creation

What’s Included

  • Message Summaries: AI-generated summaries of conversation segments
  • Message Count: Total number of messages in the archive
  • Date: The date range covered by the archive
  • Metadata: Additional context (JSON format, optional)

What’s Not Included

Archives contain summaries, not the original messages. The original message text is not preserved in archives for privacy and storage efficiency.
Read-Only Access: This MCP server only provides read access to archives. Archives cannot be created, modified, or deleted via these tools.

Integration with Time Utilities

The Conversation Archives server works seamlessly with the Time Utilities MCP server for date-based filtering:

Example: “Show me last month’s archives”

1. Time: relative-time("last month")
   → Returns: start_date: "2026-01-01", end_date: "2026-01-31"

2. list-archives(start_date: "2026-01-01", end_date: "2026-01-31")
   → Returns: All archives from January 2026

3. get-archive(date: "2026-01-15")
   → Returns: Full summary for January 15, 2026

Example: “What did we discuss 3 days ago?”

1. Time: relative-time("3 days ago")
   → Returns: "2026-02-01"

2. get-archive(date: "2026-02-01")
   → Returns: Complete summary for February 1, 2026

Tips & Best Practices

  • Always use Time tools to convert natural language dates to YYYY-MM-DD format before querying archives
  • List before getting - Use list-archives first to see what’s available, then get-archive for specific summaries
  • Date format is strict - Use YYYY-MM-DD format only (e.g., “2026-02-04”, not “Feb 4, 2026” or “2/4/2026”)
  • Archives are summaries - The original message text is not preserved; only AI-generated summaries
  • Recent conversations may not yet have archives - Archives are typically generated daily
  • Use date ranges to narrow results when searching for specific topics

Date Filtering Best Practices

Good:
# Get exact date range from Time tool first
1. relative-time("last week") → start: 2026-01-27, end: 2026-02-02
2. list-archives(start_date: "2026-01-27", end_date: "2026-02-02")
Avoid:
# Don't guess dates or use non-standard formats
list-archives(start_date: "last week")  # Won't work - needs YYYY-MM-DD
list-archives(start_date: "2/1/2026")   # Won't work - wrong format

Archive Retention

Archives are retained according to the platform’s data retention policies:
  • Standard Retention: Archives are typically retained indefinitely for the lifetime of the conversation
  • Privacy: Archives are conversation-specific and only accessible with the conversation’s API key
  • No Deletion: The MCP server does not provide tools for deleting archives
Privacy Note: Archives are scoped to individual conversations. The conversation API key ensures that archives are only accessible within their original conversation context.

Error Handling

Common errors and how to handle them:

Archive Not Found

Error: “No archive found for date 2026-02-04” Cause: Either no archive exists for that date, or it hasn’t been generated yet. Solution:
  • Use list-archives to see which dates have archives
  • Check if the date is very recent (archives are generated periodically)
  • Verify the date format is correct (YYYY-MM-DD)

Invalid Date Format

Error: “Invalid date format” Cause: Date parameter is not in YYYY-MM-DD format. Solution:
  • Use Time: relative-time or today to get properly formatted dates
  • Ensure format is YYYY-MM-DD (e.g., “2026-02-04”, not “Feb 4” or “2/4/2026”)

Unauthorized

Error: “Unauthorized” Cause: Conversation API key is missing or invalid. Solution:
  • Ensure the MCP server is being called within an Agentic Loop (which provides the key automatically)
  • Verify the conversation API key is valid and hasn’t been revoked

API Reference

View the complete API specification for Conversation Archives MCP server endpoints