Chroma MCP Server

# Memory Integration with Previous Thoughts and Concepts **Description:** These rules guide how the AI should leverage previous thoughts and conceptual discussions, especially when there are no file modifications but there is valuable conceptual thinking being developed. ## When to Apply These Rules Apply these rules particularly when: - The current conversation is addressing conceptual topics without immediate code changes - The user is exploring design options, architectural patterns, or problem-solving approaches - Previous messages in the same session contain relevant context for the current question - You need to connect code changes with previous discussions about related concepts - Building on ideas that have high confidence scores from previous interactions ## How to Integrate Previous Thoughts 1. **Within Session Continuity:** - At the beginning of your response, check if this conversation continues a previous thought thread - If relevant, refer back to key insights from earlier in the conversation using language like "As we discussed earlier..." - Ensure that your response builds coherently on the previously established concepts - Consider confidence scores from previous interactions when deciding which threads to continue 2. **Sequential Thinking Tools:** - For complex reasoning tasks spanning multiple messages, use the `mcp_chroma_dev_chroma_sequential_thinking` tool to record the thought process - Structure thoughts with clear numbering: "Thought 1 of N", "Thought 2 of N", etc. - Include a logical progression that explicitly connects to previous thoughts in the sequence - Reference code contexts from related discussions when building on file-modifying interactions 3. **When Providing Conceptual Guidance:** - Use the `mcp_chroma_dev_chroma_find_similar_thoughts` tool to search for relevant previous thoughts - Reference similar past considerations with "This relates to our previous discussion about X..." - Cross-reference related concepts that emerged in earlier parts of the conversation - When referring to code, cite related snippets that were captured in previous interactions 4. **Knowledge Integration Strategy:** - Synthesize knowledge across related conversations by using the `session_id` parameter to retrieve thoughts within the same session - Build a coherent mental model by explicitly connecting new concepts to previously established ones - When appropriate, use the phrase "Building on our previous discussion..." to signal continuity - Leverage bi-directional linking between code and discussions to connect concepts with implementations 5. **Documentation of Conceptual Evolution:** - For important conceptual breakthroughs, mark them with "Key insight:" to make them easier to identify later - When recording sequential thoughts, ensure the `next_thought_needed` parameter is set appropriately to indicate if further development is expected - Use consistent terminology across responses to maintain conceptual integrity - For high-confidence insights that lead to code changes, note the relationship to potentially create derived learnings 6. **Code Context Connection:** - When discussing code that has been modified in previous interactions, reference the relevant context - Use tool sequence patterns to understand the evolution of solutions across conversations - Connect conceptual discussions to concrete implementations where applicable - Leverage bidirectional links between chat history and code chunks when available ## Implementation Details When using the sequential thinking tools: ```python # Example of sequential thinking tool use mcp_chroma_dev_chroma_sequential_thinking( thought="[Content of the current thought]", thought_number=current_number, total_thoughts=expected_total, session_id="[current_session_id]", branch_id="[optional_branch_id]", branch_from_thought=0, # Set to parent thought number if branching next_thought_needed=True # Whether more thoughts are expected ) ``` When finding similar thoughts: ```python # Example of finding similar thoughts mcp_chroma_dev_chroma_find_similar_thoughts( query="[Current topic or question]", session_id="[current_session_id]", # Optional: limit to current session n_results=5, threshold=0.7, include_branches=True ) ``` When accessing context-rich chat history: ```python # Example of finding related code contexts mcp_chroma_dev_chroma_query_documents( collection_name="chat_history_v1", query_texts=["[Related concept or implementation detail]"], n_results=3 ) ```