Skip to main content
Glama

Server Configuration

Describes the environment variables required to run the server.

NameRequiredDescriptionDefault
WILLOW_PG_DBNoPostgres database namewillow
SAP_SAFE_ROOTNoSAFE folder root~/.sap/Applications
WILLOW_PG_USERNoPostgres user (Unix socket auth)$USER
WILLOW_STORE_ROOTNoSQLite store directory~/.willow/store
SAP_PGP_FINGERPRINTNoPinned GPG fingerprint

Instructions

Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.

Willow sovereign agent platform (willow-mcp). FIRST CALL of every session: session_enter(app_id, session_id) — it returns your entry_mode and, for a dispatched specialist, your assignment. Read the bundled 'session-start' skill for the full lifecycle. app_id=willow is the human-orchestrator seat (WILLOW_HUMAN_ORCHESTRATOR=1); agents must use their OWN app_id, never willow, and never pass dispatch_id as willow. Close a session with session_handoff_write (human path) or handoff_write_v4 (dispatch path). Core verbs: store/get/search records; kb_ingest / kb_search the knowledge base; task_submit + task_status for sandboxed Kart execution. Kart tasks are network-isolated by default: egress needs THREE keys held at once — the task_net capability in your manifest, the operator's consent.internet, and an unexpired operator-issued lease. Ask with lease_request; the operator seals the pair and net_authority_drain mints it (or the operator runs willow-mcp grant-net). No tool mints a lease without the operator's seal. Pass app_id on every call — it matches your manifest in $WILLOW_HOME/mcp_apps/<app_id>/manifest.json. tools/list for the willow desk advertises desk_core (≤50 verbs), not the full callable ACL; set WILLOW_MCP_ADVERTISE=full to restore the full listing.

Capabilities

Features and capabilities supported by this server

Protocol revision2025-11-25

CapabilityDetails
tools
{
  "listChanged": true
}
prompts
{
  "listChanged": false
}
resources
{
  "subscribe": false,
  "listChanged": false
}

Tools

Functions exposed to the LLM to take actions

NameDescription

No tools

Prompts

Interactive templates invoked by user choice

NameDescription

No prompts

Resources

Contextual data attached and managed by the client

NameDescription
store_collectionsList every SOIL collection visible to this server instance. Returns collection names and count — the same data store_collections exposes, as a static MCP resource.

TDQS

B3.3/5.0

Scored across 147 tools

Disambiguation4/5

The tool set is enormous but organized into clear prefix families (store_*, grove_*, gap_*, knowledge_*, kb_*, dispatch_*, envelope_*, session_*, fork_*, etc.) with detailed descriptions. A few pairs could be confused (e.g., store_search vs store_search_all, grove_send_message vs grove_bus_send, handoff_write_v4 vs session_handoff_write) but their scopes are distinct enough that an agent reading descriptions would likely pick correctly.

Naming Consistency5/5

All tool names follow a consistent snake_case verb_noun (or prefix_verb_noun) pattern. Prefixes denote the domain (store, grove, gap, knowledge, kb, dispatch, fork, envelope, session, human, etc.), and verbs are consistently descriptive. There are no mixed conventions like camelCase or abbreviated ambiguities.

Tool Count1/5

147 tools is extreme for any MCP server, even one covering a broad governance/fleet management domain. The sheer volume makes the surface area unwieldy and increases the risk of a tool being overlooked or mis-selected. Per the calibration, 50+ tools is a strong mismatch; 147 far exceeds that threshold.

Completeness5/5

The tool set provides comprehensive coverage across its apparent domains: store CRUD and search, messaging and fleet coordination, knowledge base lifecycle, gap/backlog management, task queueing, dispatch workflow, fork management, envelope governance, session binding/reconciliation, human-required queue, and attestations. Each domain has create/read/update/delete or lifecycle operations, leaving very few obvious gaps.

Maintenance

ActivityActive
ResponsivenessNo recent community issues