com.getfreedomos/freedom-mcp

Mark a pending attention directive as acked after the host session has taken the instruction. Use when YOU are Grok/Claude/a host builder and you just executed (or deliberately skipped) a directive you polled — for the same operator who owns the queue. Idempotent on already-acked → not found. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Add ONE activity to an agent's activity plan without regenerating the whole plan. Use to give an agent a new recurring or one-off deliverable. (To rebuild the entire plan, use recalibrate_agent_jd with regenerate_activities=true instead.) [sensitive-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Track a personal commitment, deadline, birthday, appointment, or obligation. ALWAYS use this (not save_knowledge) when the user mentions: birthdays, due dates, deadlines, tax filings, events to plan, gifts to send, things they need to do by a certain date, or anything they want reminded about. Works across all life domains (work, personal, family, home). For supporting context (e.g. gift ideas, who the person is), pair this with save_knowledge scope="personal". [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Store one piece of REAL Customer Evidence for this company (paying-customer words/behavior, telemetry, review, operator-relayed quote, prospect signal, or agent-as-user). Evidence outranks generated ICP simulation. Use when the operator pastes a real customer quote, a call note, a review, or a provenanced usage signal — NOT for inventing personas (use Customer Hunter / create_icp for hypotheses). [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Add a new lead to the Leads CRM (crm_leads) — the table the Leads tab, triage, and outreach all use. Idempotent on (company, email) when an email is given. Provide at least an email OR a name. The lead appears on the Leads tab and is auto-triaged. Routing: CRM/sales → add a lead or prospect → use this [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Gather comprehensive team and company context for talent strategy analysis. Returns current team composition, growth signals, capability gaps, and integration status. Use when the user asks "what roles am I missing?", "who should I hire next?", "analyze my team", or "what gaps does my team have?". YOU are the strategist — this tool gathers the data, YOU provide the PhD-level talent recommendations.

Append one durable speech/taste preference for THIS operator only (re-injected on their next voice session mint). Use when they say something was hard to follow, how cards should sound, or "remember I prefer…". For this user_id only — does not edit the shared FreedomOS CoS template. Apply the note in the current call when you can. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Append rows to a Google Spreadsheet. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Approve a content item for publishing — or REJECT it with approved:false. Use when user says "approve it", "looks good", "publish that" (approve), or "reject it", "drop that duplicate", "don't publish" (approved:false). [sensitive-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Archive (or restore) a content pipeline — flips is_active off/on, mirroring the Content Pipeline UI's soft-delete/restore. No data is deleted or cascaded. Use when the user says "archive this pipeline", "pause my newsletter automation", "turn off this pipeline", or "bring back my archived pipeline" (pass restore:true). [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Attach an existing freedom-ai GitHub PR URL to a product request and **resolve it by construction** (card → approved, product_status=fixed, history comment, filer resolution notify). Use when you (or a coding agent) opened a real PR for the fix — no separate human close step. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Audit whether Freedom OS appears in AI-generated search results. Sends a search query to external LLMs (Claude, Grok, Gemini, Perplexity) and checks each response for brand mentions. This is a competitive SEO/GEO auditing tool — like a mystery shopper for AI search engines. It does NOT answer questions or delegate work.

Perform batch operations on a Google Spreadsheet (formatting, merging, etc.). [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Browse a web page in a real browser and take a screenshot. Returns page content and a screenshot image. Use when you need to SEE what a page looks like (visual audit, brand check), interact with JavaScript-heavy pages, or capture visual evidence. The screenshot is returned as an image you can analyze directly with your vision. [sensitive-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Cancel a pending attention directive (operator changed mind / wrong target). Use when the operator says drop/cancel that instruction to Grok or Claude, or CoS realizes the target_session_id was wrong — for THIS operator only. Does not reverse work the host already did. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Cancel a commitment without completing it — marks it cancelled. Use when the user says "cancel that", "never mind, drop it", or "that's not happening anymore" for something already tracked. For finished work, use complete_commitment instead. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Capture an idea into the user's Ideas. Use when user shares an idea they want to save for later. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Run your deliverable past the company's customer truth: REAL Customer Evidence first (when stored), then generated ICP as labeled simulation. Returns honest feedback — what would make them engage, scroll past, or what's missing. Use on customer-impact deliverables before sending.

Check your own agent email inbox (receive-only) for messages sent to your @agents.getfreedomos.com address, and read them. Returns recent unread messages: sender, subject, a safe text snippet, any OTP codes, and login links that are safe to open — reads only YOUR mailbox. Use when a login, sign-in, or verification flow tells you it emailed a one-time code or magic link to your agent address and you need to retrieve it. [sensitive-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Mint a paste-ready Builder claim recipe for a FreedomOS product request so a host coding agent (Grok Build / Claude Code) with Harness + gstack can implement the class fix. Pins the FreedomOS frontier coding model (TIER_ROLES.frontier). Does NOT run the coding agent or open a PR by itself — use after product team accepted the request. FreedomOS product-inbox members only. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Reset all learnings for a pipeline and start fresh. Use when user says "forget what you learned", "start fresh with the style", "reset the learnings", or "clear the feedback history". [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Mark a commitment as completed. Use when the user says they finished something or a deadline has passed. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Create or update a widget on your agent dashboard. Use this to display key metrics, charts, tables, or timelines that help the user understand your work at a glance. Each call creates or updates one widget. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Confirm a pending MCP capability approval by spoken (or chat) yes/no. Pass approval_id from the approval_required tool result. decision: approve | reject | later. Runs the SAME process-approval pipeline as tapping Approve on the card — does not bypass integrity rails. Use on voice when the operator says approve/yes or reject/no after a capability ask. Do NOT invent an approval_id. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Queue a short instruction for an external coding/builder session (Grok terminal, Claude Code, or future FreedomOS runtime). Does NOT type into their UI — the session must poll FreedomOS (poll-fo-directives.sh or list_attention_directives) and act. Use when the operator says "tell Grok…", "have Claude…", or CoS should route reversible work off the call. Pass the same target_session_id the host polls (e.g. grok-<id>, claude-<id>). [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Add a new feature to the Feature Index. Use when user says "I built X", "add feature Y", "track this capability", or describes a product feature they want to market. Features can later be pushed to Content Pipeline for marketing content. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Create a folder in the knowledge base for organizing files. Folders can be nested (e.g., "partners/acme"). Use for deal rooms, topic grouping, or any organizational structure. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Create a new Google Doc in the user's Freedom OS folder. Use for JDs, deliverables, and shared documents. By default, creates beautifully formatted docs. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Create a NEW Ideal Customer Profile (ICP) from scratch and save it — no Customer Hunter UI needed. Use this when get_icps returns hasICPs:false (the company has none yet) or to add another target customer profile. To CHANGE an existing ICP, use update_icp instead. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Add a key result to an objective (the KR in OKR). Key results are measurable outcomes that track progress toward the objective. You can identify the parent objective by title or ID. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Initialize a new multi-step project with a persistent Master Plan artifact. Call this BEFORE starting any complex, multi-tool task. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Create a complete Meta (Facebook/Instagram) ad draft — campaign + ad set + creative + ad — ALL in PAUSED state, spending nothing. Use when the user wants to set up or draft an ad. Activation is a separate human-approved step (set_meta_ad_status). [sensitive-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Create a new objective (the O in OKR). Objectives are aspirational goals. After creating one, use generate_key_results to get intelligent, context-aware key result suggestions, then create_key_result to add the best ones. An objective without key results has no way to measure progress. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Create a new content pipeline to automate content creation. Use when user says "set up a changelog", "create a newsletter pipeline", "send team updates", "automate my X posts", or describes input→output automation. Output types: changelog (public product updates), team_update (internal team email via Freedom OS), report (email to specific recipients), customer_newsletter (external customers - requires user Email MCP like Mailchimp), social_post (x/linkedin/instagram/facebook/threads via the gated publish owner). [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Draft a Play (growth_tactics with steps + human review) from an oversized agent activity. Does not run the play — operator Agrees on Focus first. Use when a run hit step/continuation limits because the work is multi-unit. [sensitive-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Create a new Google Spreadsheet with optional headers. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Create a new growth tactic for the company. Use when user wants to add a tactic, strategy, or action item. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Deactivate (archive) an AI agent/specialist from the team. Use when user says "remove [agent]", "deactivate [agent]", "archive [agent]", "fire [agent]", "delete [agent]". The agent is soft-deleted (is_active=false) and can be reactivated later. Cannot deactivate Linnet (the orchestrator). [sensitive-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Approve or deny a Command Center card. This processes the decision through the full approval pipeline including trust scoring, autopilot evaluation, skill learning, and deliverable queue progression. [sensitive-tier, initiates a multi-step agent process — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Delete a saved Ideal Customer Profile (ICP). Mirrors the Customer Hunter UI's delete: deactivates any reviewer agent built from this ICP, strips it from every content pipeline that targets it, then ARCHIVES (does not permanently remove) the ICP file. Use when the user says "delete this ICP", "remove this customer profile", or "get rid of this persona". [sensitive-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Delete an idea from Ideas. Can identify by content snippet, ID, or "newest"/"latest". [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Archive a key result (safe delete — recoverable, never hard-deleted). The KR is moved out of the objective's live list into a recoverable archive. Identify by title (preferred) or ID; optionally scope by parent objective. If the title is ambiguous it refuses and lists the matches — pass an ID to disambiguate. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Archive a knowledge file by slug (soft delete). The file is moved to _archived/ and can be restored later. Use when the user explicitly asks to remove a knowledge document. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Archive an objective and its key results (safe delete — recoverable, never hard-deleted). Identify by title (preferred) or ID. If the title matches more than one objective it refuses and lists them — pass an ID to disambiguate. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Archive a tactic (safe delete - recoverable). Can identify by title instead of ID. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Run an adversarial deliberation on a decision. Multiple AI perspectives argue opposing positions over multiple rounds, iteratively strengthening arguments, and converge on a recommendation with confidence scoring. Use for important decisions where you want to stress-test options from multiple angles. Over MCP the deliberation runs in the background: the first call returns a run_id immediately; call deliberate again with { run_id } (plus the same companyId) after ~1-2 minutes to fetch the result. Routing: Important decision → deliberate for adversarial analysis

Scan the company's connected source code (its GitHub repo, via the Pulse connection in Smart Tools) and DRAFT a capability list — shipped FEATURES (each citing the file that proves it) plus attempted can't-do LIMITS — for the operator to ratify. It writes NOTHING: only items the operator ratifies become authoritative capability the marketing agents and the Integrity Gate use. Read-only; never executes or sends code. If no repo is connected it tells the operator to connect one in Pulse first. Use to populate or refresh a software product's capabilities without hand-maintaining them. Routing: Operator wants to pull their product's real features from its code (instead of typing them) → use this [sensitive-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

SPIKE tool (.agent/design-docs/2026-07-09-magical-onboarding-buildout-mode.md, "The Assignment"). Give it a business's public URL; it reads the page and returns a derived ICP, a derived brand voice, and 3 ready-to-post drafts written in that voice — the onboarding lead magnet's engine, spike-grade. Nothing is persisted to any company. Use when your operator asks you to run the website-derivation spike against a real business URL, to show the operator what an agent team can already see about their business. [sensitive-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Produce two outreach draft variants (A/B) for a lead given an angle. Both drafts are warm and kind by design (P10) — variants differ in angle of helpfulness (subject hook, opening framing, call-to-action) not in tone. Drafts are written to lead_drafts as pending_review. Returns IDs + previews. Use after synthesize_lead_hypothesis to draft initial outreach. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Draft a company tenet (mission or vision) FROM the company's existing website, for the operator to ratify or edit — instead of asking them to type it into a blank field. Use when a tenet is empty but the company already exists (has a website). Returns a DRAFT proposal with evidence and a confidence level; it writes NOTHING — the operator authors by confirming (Slice-3 update_company). The agent is a mirror, not an author: the draft is grounded in the site, never invented. [sensitive-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]