Atlas / MCP servers / circuitwall / Jarela

JarelaBLOCK

Name: Jarela
Author: circuitwall

mcp/circuitwall/jarela

MCP server by circuitwall · Audited 2026-10-12

⚑ Report / re-scan Source ↗

Jarela is a lightweight, full‐featured multi‐agent application that runs locally on your machine.

Verdict

BLOCK

Grade

Trust score

22 /100

Exposed tools

200 155r · 78w · 29d

Transport

streamable-http

License

Apache-2.0

Stars

Overview

From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.

Jarela is a local-first desktop UI for orchestrating LangGraph agents on your own machine. It keeps agent state, memory, schedules, and model/tool configuration in SQLite under a local data directory, with no hosted backend required.

Highlights

Local-first, single-process Next.js app
Multi-provider LLM support: Anthropic, OpenAI, Google GenAI, Cohere, DeepSeek, GitHub Copilot
Per-agent tool policy, identity, memory, and harness configuration
SQLite-backed checkpoints, memory, schedules, proposals, and local integrations
Durable, agent-targeted lifecycle signals for restarts, approvals, configuration changes, and background work
PWA / desktop-friendly workflow with browser extension support
MCP, integration manifests, and built-in tool extension points

Quick start

npm install
npm run dev

Then open http://127.0.0.1:4312.

Documentation

Full docs are organized under the docs site and source docs folder:

Live documentation site
Docs home
Install guide
Development guide
Architecture
Extending Jarela
API reference
Contributing

Runtime and architecture

The app is built around a single Next.js process with clear layers:

app + API routes
agent runtime + prompt assembly
provider adapters
tools + MCP registry
SQLite persistence and memory stores
integrations, bridges, scheduler, and browser extension surfaces
SQLite outbox and leased signal delivery, consumed by the existing agent run queues (no additional broker)

Extension surfaces

Jarela supports explicit extension points for:

built-in LLM providers
external provider plugins
built-in tools
MCP servers
custom harnesses
integration manifests
app branding / runtime config

Use the agent tool describe_extension_surfaces and the docs guide for th

Read from source at commit 3ae0d81f68aaOBSERVED · 2026-10-12

Connect

Built from this server's own package name, version and transport as found in its source — not copied from anyone's documentation, so it cannot drift against a page we do not control. Replace the environment placeholders with a token scoped to the least it needs.

claude-code

claude mcp add jarela-oauth-proxy --env ANTHROPIC_API_KEY=${ANTHROPIC_API_KEY} --env ANTHROPIC_AUTH_TOKEN=${ANTHROPIC_AUTH_TOKEN} --env ATLASSIAN_API_TOKEN=${ATLASSIAN_API_TOKEN} --env CODEX_API_KEY=${CODEX_API_KEY} -- npx -y [email protected]

claude-desktop

{
  "mcpServers": {
    "jarela-oauth-proxy": {
      "command": "npx",
      "args": [
        "-y",
        "[email protected]"
      ],
      "env": {
        "ANTHROPIC_API_KEY": "${ANTHROPIC_API_KEY}",
        "ANTHROPIC_AUTH_TOKEN": "${ANTHROPIC_AUTH_TOKEN}",
        "ATLASSIAN_API_TOKEN": "${ATLASSIAN_API_TOKEN}",
        "CODEX_API_KEY": "${CODEX_API_KEY}"
      }
    }
  }
}

Exposed tools (200)

155 read · 78 write · 29 destructive. Blast radius: 29 tools can delete or overwrite — an agent that can be talked into calling a tool can be talked into calling this one.

Tool	Risk	Description
`API`	read	REST
`Acme`	read	d
`Default`	read	Evidence-first scaffolding: retrieval-grounded planning, anti-fabrication output discipline, and self-config proposals.
`Done`	read	completed
`Highest`	read
`Platform`	read	core infra
`browser_activate_tab`	read	Focus a browser tab by
`browser_click`	read	Click an element in the active tab. Identify the target with EXACTLY ONE of:
`browser_extract`	read	Read text or HTML from the active tab. Without
`browser_fill`	read	Type
`browser_fill_many`	read	Fill multiple fields in the active tab in one browser round-trip. Use this for forms after a
`browser_navigate`	read	Drive the user
`browser_screenshot`	read	Take a screenshot of the active tab and store it as a local file. Returns a
`browser_scroll`	read	Scroll the active tab.
`browser_snapshot`	read	Return a compact, structured map of the active tab — URL, title, headings, landmarks, and a numbered list of every interactive control (links, buttons, inputs, selects, etc.) with
`browser_tabs`	read	List browser tabs known to the Jarela extension, including tab id, window id, active/focused/pinned/foreground markers, title, URL/host when permissions allow, and whether the tab can be scripted.
`calendar_create_event`	write	Create a Google Calendar event on a calendar (default:
`calendar_delete_event`	destructive	Delete a Google Calendar event by id. Attendees are notified by default.
`calendar_get_event`	read	Fetch one Google Calendar event by id. Returns the same slim shape as
`calendar_list_calendars`	read	List the user
`calendar_list_events`	read	List events on a calendar within a time window. Defaults: primary calendar,
`calendar_update_event`	write	Patch a Google Calendar event. Only the fields you supply are changed;
`cancel_scheduled_task`	read	Cancel a previously scheduled task by id.
`cancel_watcher`	read	Cancel a previously registered watcher by id.
`check_proposal`	read	Check the status of a previously-submitted config-change proposal by its id.
`claude_delegate`	read	Delegate a feature-scoped coding task to a local Claude Code CLI process with full tool access (Read, Write, Edit, Bash, Skill, WebSearch, etc.), running inside the active workspace (call workspace_init first).
`claude_delegate_status`	read	Poll or cancel a background claude_delegate job. Agent-owned jobs return bounded status/step counts and result_key; use tool_result_get for encrypted terminal output. Legacy unowned jobs return steps/new_steps and the synchronous result shape. Use action:
`codex_delegate`	read	Delegate a focused coding task to the locally installed OpenAI Codex CLI. Call workspace_init with the target repository before this tool, or pass cwd explicitly; do not use Jarela
`compact_context`	read	Summarize and fold the earlier conversation out of the hot context so the next turns start fresh.
`confluence_add_comment`	write	Add a footer comment to a Confluence page (v2). Pass exactly one of
`confluence_add_label`	write	Add one or more labels to a Confluence page (additive — does not replace existing labels).
`confluence_create_page`	write	Create a Confluence page (v2). Pass
`confluence_delete_attachment`	destructive	Delete a Confluence attachment by id (v2). Default soft-deletes (trash, restorable). Pass
`confluence_delete_comment`	destructive	Permanently delete a Confluence comment (v2). Pass
`confluence_delete_page`	destructive	Delete a Confluence page (v2). Default soft-deletes (page goes to trash, restorable).
`confluence_get_attachment_content`	read	Fetch an attachment
`confluence_get_comments`	read	List footer comments (and inline comments by default) on a Confluence page. Returns id,
`confluence_get_labels`	read	List labels on a Confluence page (cursor-paginated). Default limit 50 (max 250).
`confluence_get_page`	read	Fetch a Confluence page by id (v2). Returns title, space_id, parent_id, version, and BOTH
`confluence_get_page_ancestors`	read	Return the parent chain (root → leaf) for a Confluence page. Useful for breadcrumbs and
`confluence_get_page_by_title`	read	Find Confluence page(s) by exact title within a space. Auto-resolves
`confluence_get_page_children`	read	List direct children of a Confluence page (cursor-paginated). Pass
`confluence_list_attachments`	read	List attachments on a Confluence page (cursor-paginated). Returns id, title, media_type,
`confluence_list_spaces`	read	List Confluence spaces (cursor-paginated). Returns id, key, name, type, status, homepage_id.
`confluence_move_page`	write	Reorder/reparent a Confluence page (v2).
`confluence_remove_label`	destructive	Remove a single label from a Confluence page. Uses the v1 endpoint (v2 Label group is still
`confluence_search`	read	Search Confluence pages with CQL (Confluence Query Language).
`confluence_update_comment`	write	Edit an existing Confluence comment (v2). Pass
`confluence_update_page`	write	Update an existing Confluence page (v2). Pass exactly one of
`confluence_upload_attachment`	write	Attach a file to a Confluence page. Pass content_base64 for binary or content_text for plain
`contract_output`	read	Synthetic output
`cred_tool`	read	needs a key
`delegate_to_agent`	read	Hand a subtask to another agent that has specialized knowledge or tools you lack.
`describe_extension_surfaces`	read	Return the curated catalog of every extension surface in this app —
`describe_provider`	write	Return capability flags (chat / invoke / stream / embed / list_models)
`documents_add_local_source`	write	Add a local folder as a Documents source so its files become searchable via documents_search.
`documents_add_remote_source`	write	Add a Jira project / JQL query / Confluence space / CQL query / GitHub repo as a document source.
`documents_index_url`	read	Fetch and index a single Jira issue, Confluence page, or GitHub PR/issue/file on demand,
`documents_list_sources`	read	List the folders Jarela is indexing for document search. Useful before calling
`documents_reindex_source`	destructive	Force an immediate incremental sync of a remote document source. Returns counts of added / updated / unchanged docs.
`documents_remove_source`	destructive	Delete a document source and all its indexed chunks. Destructive — requires
`documents_search`	read	Search indexed local documents (notes, READMEs, code, configs) by semantic similarity.
`echo`	read	Echo
`echo-async`	read	echo
`echo-stream`	read	echoes args
`file_copy`	read	Copy a file or directory. If destination is an existing directory, source is copied into it. Directories require recursive=true (default).
`file_delete`	destructive	Delete a file or directory. Non-empty directories require recursive=true. Symlinks are not followed.
`file_glob`	read	Find files by glob pattern under a directory. Prefer this over shelling out to rg --files/find/Get-ChildItem for file discovery. Supports , *, ?, and {a,b} alternation. Skips node_modules/.git/dist/build/.next/coverage/etc by default. Returns POSIX-relative paths from the search root.
`file_grep`	read	Search file contents under a directory by regex (or literal substring with literal=true). Prefer this over local_exec/terminal grep, rg, Select-String, or Get-ChildItem pipelines for code search. Optional glob filter (e.g.
`file_list`	read	List a directory
`file_mkdir`	write	Create a directory. Creates parents by default.
`file_move`	write	Move or rename a file or directory. If destination is an existing directory, source is moved into it. Handles cross-device moves via copy+remove fallback.
`file_multi_edit`	write	Apply multiple anchored str_replace edits to one file atomically. Each edit
`file_read`	read	read files
`file_stat`	read	Check whether a path exists and return its kind, size, and timestamps.
`generate_image`	read	Generate one or more images from a text prompt using Google
`generate_voice`	read	Synthesize speech with Google
`get_integration_setup`	read	Return the full setup manifest for one integration — prerequisites, ordered steps, and
`get_user_location`	read	Get the user
`github_add_comment`	write	Post a comment on an issue or pull request (GitHub treats both the same).
`github_create_issue`	write	Open a new issue in a repository. **PREFER THIS over shell-exec
`github_create_pull`	write	Open a pull request. **PREFER THIS over shell-exec
`github_create_review`	write	Submit a pull-request review (approve / request changes / leave a comment). **PREFER THIS over
`github_get_file`	read	Read a file
`github_get_issue`	read	Fetch a single issue (or PR — same endpoint) by number. Returns body capped at 20KB, plus labels,
`github_get_pull`	read	Fetch detail on a single pull request — head/base SHAs, mergeable state, additions/deletions,
`github_get_repo`	read	Fetch repo summary (default branch, visibility, description, topics, star/fork counts,
`github_list_branches`	read	List branches in a repository with their head SHAs and protection state. **PREFER THIS over
`github_list_issue_comments`	write	List comments on an issue or PR. **PREFER THIS over shell-exec
`github_list_pull_files`	read	List files changed in a pull request, with per-file additions/deletions and (capped) patch text.
`github_list_pull_reviews`	read	List reviews submitted on a pull request (approvals, change-requests, comment-only). **PREFER THIS
`github_list_pulls`	read	List pull requests for a repository, optionally filtered by state / head branch / base branch.
`github_merge_pull`	write	Merge a pull request. **PREFER THIS over shell-exec
`github_request_reviewers`	read	Request review on a pull request from one or more users and/or teams. **PREFER THIS over
`github_search_code`	write	Search file contents across GitHub. **PREFER THIS over shell-exec
`github_search_issues`	read	Search GitHub issues AND pull requests with the same query syntax as the github.com search bar.
`github_update_issue`	write	Edit an issue or PR (same endpoint): change title, body, labels, assignees, or close/reopen via
`github_update_pull`	write	Edit a pull request: title, body, base branch, or close/reopen. **PREFER THIS over shell-exec
`gmail_create_draft`	write	Create a Gmail draft. Prefer drafts unless the user explicitly asks to send immediately.
`gmail_get_message`	read	Fetch one Gmail message by id, including full headers and the decoded text body (text/plain
`gmail_list_labels`	read	List all Gmail labels (system + user). System label ids are uppercase (INBOX, UNREAD,
`gmail_modify_message`	destructive	Add or remove labels on a Gmail message. Common patterns:
`gmail_search`	read	Search the user
`gmail_send_email`	write	send mail
`gmail_trash_message`	destructive	Move a Gmail message to Trash (reversible for 30 days, not permanent delete). Only call this
`icloud_calendar_create_event`	write	Create an event on an iCloud calendar.
`icloud_calendar_delete_event`	destructive	Delete an event by UID. For a recurring series this removes every
`icloud_calendar_get_event`	read	Fetch a single event by UID, including the raw iCalendar source.
`icloud_calendar_list_calendars`	read	List iCloud event calendars (VEVENT collections). Returns the URL
`icloud_calendar_list_events`	read	List events in a calendar between two ISO timestamps. The calendar
`icloud_calendar_update_event`	write	Patch fields on an existing event. Only the supplied fields are
`icloud_mail_create_draft`	write	Save a new draft to the iCloud Drafts folder. This intentionally does
`icloud_mail_delete_message`	destructive	Move a message to the iCloud Trash (recoverable for ~30 days,
`icloud_mail_flag_message`	destructive	Add or remove standard IMAP flags on a message — typically
`icloud_mail_get_message`	read	Fetch a single message by UID. Returns headers, plain-text body,
`icloud_mail_list_folders`	read	List iCloud Mail folders (mailboxes) with message + unread counts.
`icloud_mail_list_messages`	read	Search messages in an iCloud Mail folder. Returns the most recent
`icloud_mail_move_message`	write	Move a message to a different folder. To send a message to trash,
`icloud_reminders_complete`	read	Mark a reminder (VTODO) as completed.
`icloud_reminders_create`	write	Create a reminder (VTODO) in an iCloud Reminders list.
`icloud_reminders_list`	read	List reminders (VTODO items) in a Reminders list. The calendar URL
`icloud_reminders_list_lists`	read	List iCloud Reminders lists (VTODO calendar collections). Returns
`invoke_tool`	write	Never target invoke_tool itself; call the target tool directly. Execute a permitted tool by name only after list_tools finds a tool that is permitted for this agent but was not directly loaded this turn, especially when permission_reason=
`jira_add_comment`	write	Add a comment to a Jira issue. Plain text is auto-converted to ADF (Atlassian Document Format).
`jira_add_remote_link`	write	Attach a web/external link to a Jira issue (Confluence pages, GitHub PRs, dashboards, Slack threads,
`jira_add_worklog`	write	Log time spent on a Jira issue. time_spent uses Jira
`jira_align_add_comment`	write	Post a comment on a Jira Align work item. **
`jira_align_create_dependency`	write	Link two Jira Align work items as a dependency: the predecessor must complete before the
`jira_align_create_entity`	write	Create a Jira Align hierarchy entity: program, team, release, sprint (PI), portfolio, or
`jira_align_create_item`	write	Create a Jira Align work item in a given type
`jira_align_delete_comment`	destructive	Permanently delete a comment from a Jira Align work item. Irreversible. **
`jira_align_delete_dependency`	destructive	Permanently remove a Jira Align dependency link. Irreversible. Use
`jira_align_delete_entity`	destructive	Permanently delete a Jira Align hierarchy entity (program/team/release/sprint/portfolio/
`jira_align_delete_item`	destructive	Permanently delete a Jira Align work item. Irreversible. **
`jira_align_get_entity`	read	Fetch a single Jira Align hierarchy entity by id: program, team, release, sprint (PI), portfolio,
`jira_align_get_item`	read	Fetch a Jira Align work item by id. **Must specify
`jira_align_list_children`	read	List the direct children of a Jira Align work item. **Both
`jira_align_list_comments`	read	List comments on a Jira Align work item. **
`jira_align_list_dependencies`	read	List Jira Align dependencies (links between work items) involving a given item. By default
`jira_align_list_entities`	read	List non-work-item Jira Align entities of a given type: programs, teams, releases, sprints (PIs),
`jira_align_search_items`	read	Search Jira Align work items of a given type, with optional state/owner/program/team
`jira_align_transition_item`	write	Move a Jira Align work item to a new workflow state (e.g.
`jira_align_update_comment`	write	Edit the body of an existing comment on a Jira Align work item. **
`jira_align_update_entity`	write	Patch fields on a Jira Align hierarchy entity (program/team/release/sprint/portfolio/value
`jira_align_update_item`	write	Patch fields on a Jira Align item (title, description, owner, programId, teamId,
`jira_create_component`	write	Create a component on a Jira project. assignee_type controls default assignee for issues with
`jira_create_issue`	write	Create a new Jira issue. Defaults to issue_type=
`jira_create_issues_bulk`	write	Create up to 50 Jira issues in a single API call. Each entry takes the same shape as
`jira_create_sprint`	write	Create a future sprint on a board. New sprints always start in
`jira_create_version`	write	Create a new version on a Jira project. Pass the project key — we resolve it to the numeric id.
`jira_delete_attachment`	destructive	Permanently delete an attachment from a Jira issue by id. Destructive — no undo. Look up
`jira_delete_comment`	destructive	Permanently delete a comment from a Jira issue. Destructive — no undo. Look up the id via
`jira_delete_issue`	destructive	Permanently delete a Jira issue. DESTRUCTIVE — there is no undo from the API. By default,
`jira_delete_link`	destructive	Delete an issue link (Jira-to-Jira, default) or a remote/web link. Look up the id first with
`jira_delete_sprint`	destructive	Permanently delete a sprint. Irreversible — issues are unassigned from the sprint but the
`jira_find_user`	read	Look up Jira Cloud users by email or display-name fragment. Returns accountId values
`jira_get_attachment_content`	read	Fetch a Jira issue attachment
`jira_get_board`	read	Fetch board metadata and configuration in one call: id, name, type, project_key, plus filter id,
`jira_get_changelog`	read	Fetch a Jira issue
`jira_get_comments`	read	Paginated comment list for a Jira issue. Use this when an issue has more comments than the
`jira_get_issue`	read	Fetch a single Jira issue by key (e.g.
`jira_get_project`	read	Fetch a single Jira project by key. Optionally include versions, components, and/or issue types
`jira_get_sprint`	write	Fetch a single sprint by id. Returns name, state, goal, start/end/complete dates. Use jira_search
`jira_link_issues`	write	Create an issue link between two Jira issues (Blocks, Relates, Duplicates, Cloners, etc.).
`jira_list_boards`	read	List Jira agile boards (Scrum or Kanban). Filter by project key, name fragment, or board type.
`jira_list_components`	read	List components on a Jira project. Returns id, name, description, lead, default assignee type.
`jira_list_meta`	read	List values for a Jira site-wide enum: issue types, priorities, statuses, or resolutions.
`jira_list_projects`	read	List Jira projects (paginated). Filter by name fragment via
`jira_list_sprints`	read	List sprints on a board. Filter by state (
`jira_list_versions`	read	List versions on a Jira project (paginated). Returns id, name, released/archived flags, dates.
`jira_list_worklogs`	read	List worklog entries on a Jira issue (paginated). Returns id, author, time_spent (display string +
`jira_move_issues_to_backlog`	destructive	Remove issues from their current sprint and put them back on the backlog. Up to 50 issues per
`jira_move_issues_to_sprint`	write	Move issues into a sprint. Up to 50 issues per call. Issues already in another sprint are
`jira_rank_issues`	read	Rank up to 50 issues relative to a single anchor issue (before XOR after). Pass
`jira_search`	write	Search Jira issues using JQL. **PREFER THIS over shell-exec
`jira_transition_issue`	read	Transition a Jira issue
`jira_update_comment`	write	Edit an existing comment on a Jira issue. Plain-text body is auto-converted to ADF (same as
`jira_update_issue`	write	Edit fields on an existing Jira issue: summary, description, priority, assignee, fix versions,
`jira_update_sprint`	write	Update a sprint
`jira_update_version`	write	Edit a version: rename, change dates, mark released/unreleased, mark archived/unarchived.
`jira_upload_attachment`	write	Upload a file as an attachment to a Jira issue. Pass content_base64 for binary files (PNG, PDF,
`large-async`	read	test
`large-json`	read	test
`large-text`	read	test
`large-text-flat`	read	test
`list_harnesses`	read	List available agent harness presets with ids, names, default status, builtin/custom source, and enabled sections. Read-only. Use before proposing harness changes.
`list_integrations`	write	List every integration Jarela knows how to set up, with a one-line summary and
`list_mcp_servers`	read	List every configured MCP server (stdio or http transport) with its
`list_providers`	read	List every LLM provider adapter registered in this app (built-in like
`list_reaction_scripts`	read	List the names of built-in reaction scripts available to schedule_watcher AND
`list_scheduled_tasks`	read	List every scheduled task across all agents. Each row is tagged with
`list_skills`	read	List all available skills by ID, name, and description. The same list appears in the system prompt; call this only if you need a refreshed view after writing a skill.
`list_tools`	read	List every tool currently available to this agent — built-in tools,
`list_watchers`	read	List every event-driven watcher across all agents. Each row is tagged
`mcp_echo`	read	Echo via MCP
`memory_delete`	destructive	Delete a single memory entry by namespace and key. Returns { removed: true } if a row was removed, { removed: false } if no matching row existed. Bulk deletion by namespace is not supported by this tool — list-then-delete-loop if needed.
`memory_list`	read	List memory entries, optionally filtered by namespace. With a search term, matches by meaning and by exact text in keys and values; without one, lists newest first.
`memory_read`	read	read memory
`memory_upsert`	read	Store a durable, structured memory record for proactive semantic recall. Use namespace=
`memory_write`	write	Write or update a value in long-term memory. Use this to remember facts, user preferences, or any information that should persist across conversations.

Trust audit

BLOCKgrade F · trust 22/100 Do not install this without reading the findings. The audit found something that could harm you or your machine.

Layer	What it checks	Result
L0	Provenance & inventory	FAIL
L1	Static analysis of the code	FAIL
L2	Instruction surface (what it tells the agent)	FAIL
L3	Class-specific surface	WARN
L4	Behavioural (sandbox)	SKIPPED

What the source does

Filesystem: declared (5 observation(s))
Network: declared (5 observation(s))
Shell: declared (7 observation(s))
Dependencies: not all pinned
Secrets in source: found

Findings (25)

HIGHInventory / provenance · inv.suspicious_name · CWE-1104

app/api/v1/agents/payload.ts

payload.ts

Why it matters. member named after an attack tool

Fix. remove or justify

HIGHInventory / provenance · inv.suspicious_name · CWE-1104

lib/models/editor-payload.ts

editor-payload.ts

Why it matters. member named after an attack tool

Fix. remove or justify

HIGHCode injection · code.eval_exec · CWE-78, CWE-94, CWE-95

lib/terminal/session.ts:115

exec(command: string, timeoutMs = 60_000): Promise<ExecResult> {

Why it matters. evaluates text as code

Fix. remove; use a parser or a dispatch table

HIGHCode injection · code.eval_exec · CWE-78, CWE-94, CWE-95

scripts/gmail-oauth.mjs:77

exec(cmd, (err) => { if (err) console.error("Failed to launch browser:", err.message); });

Why it matters. evaluates text as code

Fix. remove; use a parser or a dispatch table