alchemyst-mcp_skill

Overview

This skill connects your agent to the Alchemyst AI MCP server to store, retrieve, search, and inspect persistent context at https://mcp.getalchemystai.com/mcp/sse. It enables long-lived memory: documents, conversations, and structured knowledge that persist across sessions, tools, and environments. Use it whenever you need reliable context persistence, semantic search, or audits of stored knowledge.

How this skill works

The skill talks to Alchemyst’s MCP SSE endpoint using a Bearer API key passed in the Authorization header. It exposes four actions: semantic search, add/store documents, view context summaries, and list stored documents. Payloads require specific metadata shapes (note: add uses snake_case metadata, search uses camelCase) and supply user_id and organization_id to scope data.

When to use it

• Saving meeting notes, decisions, code conventions, or onboarding docs for later retrieval
• Before answering questions that may depend on project-specific knowledge—run a semantic search proactively
• Auditing or troubleshooting what an agent knows for a given user/org
• Browsing stored documents to verify content or avoid duplicates
• Building multi-session agents that must recall past interactions or resources

Best practices

• Always include the required metadata when calling add_context; missing metadata hurts retrieval quality
• Run a search before adding new content to avoid duplicates and keep the store clean
• Chunk large documents into coherent sections so each stored item is independently useful
• Version sources with dot-separated labels (e.g., project.arch.v1) rather than overwriting the same source
• Pass organization_id explicitly (use null if none) and never commit your API key to source control
• Mind the camelCase vs snake_case difference between search_context and add_context payloads

Example use cases

• Persisting architectural decisions and retrieving them during design reviews
• Storing onboarding guides and searching them when answering new hire questions
• Keeping support conversation history so agents can fetch prior threads before replying
• Recording code conventions and injecting them into code-review prompts
• Auditing stored knowledge before a release to confirm required docs are present