Reference
MCP tools
The Model Context Protocol tool catalog exposed by `seif mcp serve`.
The MCP server is built into the seif CLI (seif mcp serve) — no separate package.
Client config example
{
"mcpServers": {
"seif": {
"command": "seif",
"args": ["mcp", "serve"],
"env": {
"SEIF_ENGINE_TOKEN": "<workspace-key>"
}
}
}
}Setup guide: ../guides/mcp-setup.md.
Cold start (all MCP hosts)
On connect, seif mcp serve returns server instructions with the grounding playbook.
Call session_orient first in a session (unless your host already injected SEIF
context at launch). Supporting surfaces:
| Surface | Identifier | Role |
|---|---|---|
| Initialize | instructions | Playbook + grounding kernel |
| Resource | seif://grounding/kernel | Read the kernel without calling a tool |
| Prompt | seif_session_start | Host UI cold-start prompt |
| Tool | session_orient | First tool — grounding + store + memory |
Local (free — no engine token)
| Tool | Maps to | Description |
|---|---|---|
artifact_read | — | Read a workspace .seif artifact (SEIF-ENVELOPE-v1). Returns the parsed frontmatter and JSON body. |
artifact_validate | — | Validate a .seif file structure and integrity (local, no network). Applies the pinned per-type schema when the artifact protocol is registered. |
classification_check | seif classify | Classify text sensitivity as PUBLIC, INTERNAL, or CONFIDENTIAL using the SEIF auto-classifier (local, no network). |
context_query | seif context query | USE WHEN you need to find relevant code/docs WITHOUT reading whole files (refs first) — lexical pre-filter over structural chunks, returns unresolved ChunkRefs JSON (local, read-only). |
how | seif how | USE WHEN you need to find the right SEIF CLI verb for a natural-language intent — ranked verb-docs from the shipped capability catalog. Free, local, lexical-only. |
memory_query | seif memory query | USE WHEN you need to RECALL a past decision, 'where did we decide X', or project history — ranked memory + knowledge hits. Optional semantic blend via local Ollama; uses embedding cache when present. |
ollama_chat | — | Chat with local Ollama (free, no engine). Uses the SEIF regex should_rag_query to decide whether RAG context would be relevant; reports the decision + the live semantic index freshness (.seif/.cache/memory-embeddings/, rebuilt via seif memory reindex) so the caller can introspect. Stateless (no thread persistence); RAG decision-surface only (does not retrieve context — that lands in a future wave). |
provenance_sign | seif sign | Sign text with the local Ed25519 signing key, returning a SEIF evidence frame (local provenance, no network). Requires seif key generate first. |
provenance_verify | seif verify | Verify a SEIF evidence frame (JSON): recompute the canonical hash, check the Ed25519 signature, re-derive the evidence id (local, no network). |
relay_assignments | seif relay assignments | USE WHEN you need to see who is assigned to what in the relay — lists ACTIVE handoff assignments (executor_id + watch_slugs). Local, metadata-only. |
relay_claim | seif relay claim | USE WHEN you need to take ownership of a relay slug as an executor — claims ACTIVE ownership for an executor_id. Refuses on visible conflict unless also_watch. |
relay_close | seif relay close | USE WHEN you need to terminally finish a relay assignment or sweep stale claims — Mirror of seif relay close. |
relay_next | seif relay next | USE WHEN you need to pick up the next QUEUED task as an executor — atomically claims the oldest queued brief (executor_id: auto). Mirror of seif relay next; use when session_orient.relay.next_in_queue is set. |
relay_poll | seif relay poll | USE WHEN you need to TRACK a peer agent's progress or see what's queued/done in the relay (the answer to 'I can't see their TTY') — single-shot filtered poll; returns metadata events + next_token (mtime-based). |
relay_release | seif relay release | USE WHEN you need to hand a claimed task back to the queue — releases an ACTIVE relay claim back to dispatch. Mirror of seif relay release. |
Requires signing key for provenance_sign.
Engine-gated (workspace token required)
| Tool | Maps to | Description |
|---|---|---|
context_bootstrap | constitutional core (SEIF.md + RESONANCE.json) — after session_orient | Deep constitutional core — behavioral axioms from the embedded verified kernel + SEIF.md — use AFTER session_orient. |
context_retrieve | seif context retrieve | USE WHEN you need the actual code/doc SLICES (after context_query refs) instead of loading whole files — orchestrator retrieval loop (local chunk-refs → engine resolved slices). Prefer refs first; slices as needed. Fail-soft when engine unreachable. |
get_hygiene_status | store-v2 hygiene status | Return the STORE-v2 hygiene status (pending counts, last local check, synced snapshot) — projected from the workspace summary. Requires a workspace key. |
get_members_status | members snapshot status | Return the members snapshot status (git pins, owner trio, canonical_writer) — projected from the workspace summary. Requires a workspace key. |
get_pod | seif pod show | Get a single compose pod by its namespace id (user-slug/pod-slug), with images + members. Calls the SEIF engine — requires a workspace key. |
get_workspace_summary | engine workspace summary | Fetch the enriched workspace summary (store_v2 hygiene, members_local, sync_local, ratified config) from the SEIF engine. Requires a workspace key. |
grounding_ack | derivation self-check heuristics | Optional self-check on your first-paragraph derivation (non-blocking heuristics). Call after session_orient. |
list_pods | seif pod list | List the compose pods visible to the workspace key (with their images). Calls the SEIF engine — requires a workspace key. |
log_append | transparency log append | Submit an attestation to the SEIF transparency log, returning entry id + sequence. Requires a workspace key. |
log_get | transparency log read | Fetch a seif-log entry by id and INDEPENDENTLY verify its RFC-6962 inclusion proof + signed tree head. Requires a workspace key. |
memory_surface | seif memory-surface / MEMORY.md index | USE WHEN you need the cross-AI memory bootstrap summary (what's known across sessions). Prefer session_orient at session start. |
seif_invoke | seif gov … (governance actions) | Invoke a SEIF governance action with structured params. USE WHEN you need to: REMEMBER a durable fact (memory_add) · DEFER work for the next cycle (pending_register) · open/close a unit of work (cycle_start/close_ritual) · APPEND an amendment with no re-sign (amend) · hand off / pick up relay work (relay_brief/relay_handback/relay_next). Full action list in the action param. |
session_orient | session orient (grounding + store + memory) | Recommended first SEIF tool each session. Returns plain-conduct grounding (calibrate, verify, settle, minimal intervention, human gate) plus store snapshot and memory_surface. No derivation ritual required (pending-2026-06-08-resonance-decouple-value-from-math). |
submit_local_snapshot | seif … --push-snapshot | Submit a local CLI snapshot (hygiene/members/sync/config) to the engine — same as seif … --push-snapshot. Requires a workspace key. |
sync | seif push | Push the local .seif/ workspace state to the engine mirror + transparency log, returning the verify URL + log entry id. Requires a workspace key. |
watermark_embed | seif watermark embed | Sign a watermark-embed provenance envelope for an (input, output) audio file pair via the SEIF engine. Requires a workspace key. |
watermark_extract | seif watermark extract | Look up the signed watermark envelope for a recovered audio file via the SEIF engine. Requires a workspace key. |
Pass token via client env SEIF_ENGINE_TOKEN or MCP serve flags --token / --token-file.