# Kapella Hub Firekeep MCP server

Source: https://glimind.com/server/mcp-registry%2Fio.github.kapella-hub%2Ffirekeep
Registry id: `mcp-registry/io.github.kapella-hub/firekeep`
Status: unknown

## Install

```json
{
  "mcpServers": {
    "kapella-hub-firekeep-mcp-server": {
      "command": "uvx",
      "args": [
        "firekeep-client"
      ]
    }
  }
}
```

Claude Code:

```sh
claude mcp add kapella-hub-firekeep-mcp-server -- uvx firekeep-client
```

## Tools (75)

- `action_after`
- `action_before`
- `audit_memory`: Memory access audit trail from replay events. Use to investigate stale memory or verify provenance. Args: action: "read" or "write" (default: both). memory_chai
- `corpus_delete`: Delete an ingested source and every chunk of it. Irreversible. Args: source_name: Source name as shown by corpus_sources.
- `corpus_ingest`: Ingest a business document (wiki, Jira, API docs, SOP) into the team corpus. Chunked and embedded; every chunk is searchable via memory_recall. For your own act
- `corpus_sources`: List ingested business documents with their chunk counts.
- `ctx_abandon_session`: Abandon a session without saving to long-term memory. Use this for sessions started by mistake or no longer relevant. Args: session_id: Session to abandon (defa
- `ctx_complete_session`
- `ctx_get_shadow`
- `ctx_list_sessions`: List your recent sessions. Only sessions that belong to you (your verified identity) are listed; agent_id narrows that list, it never widens it. Args: status: F
- `ctx_resume_session`
- `ctx_start_session`
- `ctx_update`: Update your working context with new information. Call this as you work to record your plan, decisions, file knowledge, and progress. This data persists across 
- `decision_board`
- `decision_board_check`: Collect answers from a Decision Board started earlier via ``decision_board``. Use the ``board_id`` returned by ``decision_board``'s pending response. Returns th
- `eval_session`: Eval metrics for a completed session (tool success rate, failure rate, duration, etc). Auto-computed on session completion, available immediately after. Args: s
- `eval_summary`: Aggregate eval metrics across recent sessions (avg scores, failure rates, trends). Args: limit: Sessions to aggregate (default 20).
- `hands_act`
- `hands_browser`
- `hands_find`
- `hands_observe`
- `hands_request_permit`
- `hands_status`
- `hands_task_end`
- `hands_task_start`
- `knowledge_ingest`
- `knowledge_ingest_url`: Crawl a URL (SSRF-guarded, bounded depth/pages) and ingest each fetched page through the docs->skills front door — same pipeline as knowledge_ingest, but the co
- `memory_feedback`
- `memory_handoff`: Generate a team handoff summary for a project: what was done, open items, where to pick up. Fetches contributor activity and recent memories, then synthesizes a
- `memory_health`: Check FirekeepCortex connectivity (Neo4j, Qdrant, Redis). Use when other memory tools return errors.
- `memory_learn`
- `memory_recall`
- `memory_stream`: Queue a raw event for background LLM extraction (Sleep Cycle worker). For high-volume raw events (CI logs, IDE events, monitoring). If you already know the stru
- `relay_broadcast`: Send a message to a channel for real-time coordination. Messages are delivered to active listeners and stored in a backlog buffer so late-joining agents can cat
- `relay_claim`: Claim a resource (file/task) to prevent duplicate work. Uses atomic locking. If the resource is already claimed, returns info about who holds it and when it exp
- `relay_deregister`: Remove your presence entry. Call on session exit. Args: agent_id: Your agent identifier
- `relay_get_dm`: Direct messages for an agent (newest first). Returns only messages addressed to YOUR member (the member the label was registered to when each was sent); another
- `relay_get_messages`: Get recent messages from a channel backlog (poll-based, not real-time). Returns messages stored in the channel's backlog buffer, newest first. Args: channel: Ch
- `relay_heartbeat`: Extend a lease's TTL. Only succeeds if fencing_token matches. Call this periodically to keep your lease alive. If you stop heartbeating, the lease expires autom
- `relay_heartbeat_presence`: Refresh your presence TTL. Call on every user turn. Also used to backfill session_id and goal after ctx_start_session. Args: agent_id: Your agent identifier ses
- `relay_lease`: Acquire a lease on a resource with a fencing token. Leases are like claims but with stale-writer protection. The returned fencing_token must be passed to relay_
- `relay_lease_status`: Check the lease status of a resource. Returns who holds the lease, their fencing token, time remaining, and how many agents are waiting in the queue. Args: reso
- `relay_post`: Post to the persistent bulletin board. Bulletin posts persist for a configurable TTL and are visible to all agents across all sessions. Use for cross-session co
- `relay_read`: Read the bulletin board, optionally filtered by tags or author. Args: tags: Filter by any matching tag author: Filter by author limit: Maximum posts to return
- `relay_register`: Register this agent as online in the presence registry. Call this at session start (briefing hook). Status is computed as 'active' (heartbeat within 10 minutes)
- `relay_release`: Release a lease or legacy claim resource. Only the holder can release. Args: resource_id: The leased/claimed resource to release agent_id: Your agent identifier
- `relay_send_dm`: Send a direct message to another agent or to the dashboard. Messages appear in the recipient's DM inbox and can be polled via relay_get_dm or the dashboard DM d
- `relay_status`: Get relay system status: active channels, bulletin count, active claims. Use this to understand the current state of agent coordination.
- `relay_task_delete`: Delete a task permanently from the queue. Use this to clean up completed, cancelled, or stale tasks that are no longer needed. Args: task_id: The task ID to del
- `relay_task_list`: List tasks, optionally filtered and ordered oldest-first. Use this to check your inbox (filter by your agent ID) or see all pending work across agents. Each tas
- `relay_task_post`
- `relay_task_update`: Update a task's status or reassign it. Call this when you start working on a task (status="in-progress"), finish it (status="completed"), or need to hand it off
- `relay_who_is_online`: List all registered agents with computed status. Status is computed from heartbeat recency: - "active" = heartbeat within last 10 minutes - "idle" = registered 
- `replay_context_at`: Reconstruct the agent's context at an event (nearest snapshot if no exact match). Args: session_id: Session containing the event. event_id: Event to inspect con
- `replay_inspect`: Full details of a single trace event. Args: event_id: Trace event ID. brief: If True, summarize payload as one-line-per-field (saves tokens when you only need s
- `replay_narrow`: Rank likely contributing events for a failure (walks trace links backward). Use after finding a failure event to identify root-cause candidates — stale memory, 
- `replay_summary`: Session trace summary: event counts by type, duration, failures, agents. Args: session_id: Session to summarize.
- `replay_timeline`
- `runbook_ack`
- `scope_ask`
- `scope_check`: Check a scope session for answered and still-pending screens. Args: scope_id: Session ID returned by scope_start/scope_ask/scope_post
- `scope_complete`: Mark a scope session completed (D-S22). Call this when you're done with a scoping session, so it doesn't sit active for up to 72h before the abandonment sweep c
- `scope_post`: Post a non-gating scoping screen and return immediately. Use during execution when a question shouldn't block progress. Poll for the answer later with scope_che
- `scope_start`: Start a FirekeepScope scope-clarification session. Call this once per scoping conversation when your agent runtime has no local FirekeepScope companion (headles
- `sentinel_get_events`: Get recent environment events observed by FirekeepSentinel. Call this to check what has changed in your environment — container state changes, new git commits, 
- `sentinel_get_health`: Get health status of FirekeepSentinel — Redis connectivity, collector status, and Docker container states. Call this to quickly check if Sentinel is operational
- `sentinel_push_event`
- `skill_add_step_specs`
- `skill_create`
- `skill_list`: List existing skills. Use to check before creating a duplicate. Args: status: Filter by lifecycle state: active, draft, deprecated. project: Optional project sc
- `skill_recall`: Retrieve relevant active (and trial) skills for the current task. Call at session start or when approaching a complex problem to surface team-shared procedures 
- `vault_delete`: Delete a secret from the vault (irreversible). Args: key: Secret name.
- `vault_list`: List vault secrets (metadata only — names, categories, tags; never values). Args: category: Optional filter (e.g. "ssh", "api_key").
- `vault_retrieve`: Retrieve and decrypt a secret from the vault. Args: key: Secret name.
- `vault_store`

Measured by Glimind (https://glimind.com): MCP handshake + tool list only, no tool is ever called. CC BY 4.0.