Reference

The eight tools

Exposed over stdio by mcp_server/server.py. Plus a ninth helper, tuskpoint_info, that returns ready-to-paste client configuration so an agent can wire itself up.

Write

Persist new state or branch a run.

PUTcheckpoint_save(thread_id, state_json)

Serialize agent state, gzip it, and store it as an immutable Walrus blob. Also writes a one-line summary to MemWal.

→ { checkpoint_id, blob_id, thread_id }

e.g. checkpoint_save("run-42", '{"step":"research","sources":["a","b"]}')

FORKcheckpoint_fork(source_thread, source_id, new_thread)

Git-branch an agent run. Copy any checkpoint into a new thread to replay a different path — the original stays untouched.

→ { new_thread_id, checkpoint_id, blob_id, forked_from }

e.g. checkpoint_fork("run-42", "0c3b84d1-…", "run-42-alt")

Read

Deterministic, content-addressed reads from Walrus.

GETcheckpoint_load(thread_id, checkpoint_id?)

Load a specific checkpoint by ID (or the latest). Deterministic, content-addressed read straight from Walrus.

→ { state, checkpoint_id, blob_id }

e.g. checkpoint_load("run-42", "0c3b84d1-…")

LScheckpoint_list(thread_id)

List every checkpoint for a thread, newest first, with lineage, timestamps, and summaries from the manifest.

→ { count, checkpoints[] }

e.g. checkpoint_list("run-42")

REScheckpoint_resume(thread_id)

Rehydrate the latest state so an agent can continue exactly where it left off after a crash.

→ { state, checkpoint_id }

e.g. checkpoint_resume("run-42")

DIFFcheckpoint_diff(thread_id, id_a, id_b)

Compare two checkpoints and report exactly what was added, removed, or changed between them.

→ { added, removed, changed, human_readable }

e.g. checkpoint_diff("run-42", id_a, id_b)

AUDITverify_trail(thread_id)

Audit a thread end-to-end. Re-fetches every content-addressed blob so tampering or corruption shows up as a failed step.

→ { ok, checkpoint_count, verified, steps[] }

e.g. verify_trail("run-42")

Discover

Find the right checkpoint in plain English.

FINDcheckpoint_search(query)

Ask a question in plain English. MemWal returns the nearest checkpoint summaries — pointers you then load exactly.

→ { results: [{ text, distance }] }

e.g. checkpoint_search("when did the writer start?")

tuskpoint_info

Calling tuskpoint_info returns the full tool list plus copy-paste MCP config for Claude Desktop, Claude Code, Cursor, Windsurf, and the generic .mcp.json form — so an agent can discover how to connect without leaving the chat.