MCP tools
Every tool returns Markdown, which models read better than JSON. Errors come back as tool errors with a readable message.
Reading
engram_session_start
Opens a session for a goal and returns the context pack. goal? · agent? · files? · budget?
engram_context
A briefing for a task within a token budget: principles, relevant decisions, pitfalls, the last session. task? · files? · budget? (200 to 20000)
engram_recall
Full-text search with filters. query? · kinds? · tags? · files? · include_inactive? · limit? (up to 50)
engram_get
One memory in full, with its links. id
engram_related
Walks the graph around a memory in both directions. id · depth? (1 to 4) · relations?
engram_why
A timeline for a topic: decisions, what they replaced, what was rejected and why, related pitfalls. topic
engram_check
Checks a plan against pitfalls and previously rejected approaches. plan · files?
engram_summary
A project overview from memory. detail?: short or full
Writing
engram_remember
Stores a memory. If an active memory has the same kind and title, it is updated instead. kind · title · content? · tags? · files? · importance? (1 to 5) · status? · supersedes? · links?: [{ to, relation, note? }]
engram_update
Edits a memory. id plus any of title, content, kind, tags, files, importance, status
engram_link
Connects two memories. from · to · relation · note?
engram_forget
Archives a memory, or deletes it with hard. id · hard?
engram_session_end
Closes the session with a handoff summary. summary · session_id? · tags?
engram_index
Runs the ADR and git indexer again.
Resources
engram://summary: the project summaryengram://memory/{id}: a single memory (the list is browsable)
Prompts
Each bundled skill is also an MCP prompt, with an optional input argument.