Skip to main content
Glama

Send a typed intent to another IC member's agent inbox

ic_agent_inbox_send_envelope

Route a typed intent (ping / request_meeting / send_intro) to another IC member's agent inbox. Recipient's policy engine decides what happens — store + notify, store + queue-for-tap, silently drop (blocklist), or refuse (inbox closed). Server-side: sanitizes body (C0 controls / zero-width / NFKC), wraps into the per-intent payload, persists thread + envelope + sender-history + audit, evaluates policy, returns the decision. Scopes per intent: ping → agent:ping (ai-floor+); request_meeting → agent:request_meeting (ic-member+); send_intro → agent:send_intro (ic-member+). send_intro brokers an introduction TO the recipient and requires intro_target_name + body (the intro_pitch) + expected_outcome + consent_target_has_opted_in=true (anti-spam — you MUST have the target's consent). Idempotency: pass idempotency_key to make the (token, key) pair cached for 24h. Returns: { ok, envelope_id, thread_id, state, policy_decision }. Recipient inbox closed → mcpError. Blocklisted senders get an opaque ok-shape with random ids (silent-block — no persistence visible to the sender; the audit row is server-side only). PRECHECK: call ic_agent_directory_lookup first — a member whose inbox_status is "closed" (the default for newly-joined members) cannot be reached and this verb will refuse. v1 SHIP note: request_meeting wraps body into context_summary with sensible defaults until the agent-console UI exposes full per-intent args.

Input Schema

TableJSON Schema
NameRequiredDescriptionDefault
toYesRecipient IC member slug (e.g. 'nicholas-e', 'michalis'). Resolve the slug via ic_agent_directory_lookup and check `addressable_by_you` on the hit, NOT `inbox_status`: inbox_status is the recipient's posture, and an open inbox can still refuse you because the send-side intent scopes are tier-gated. If addressable_by_you is false, that hit's blocked_intents names the scope you are missing.
bodyNoFree-form context. REQUIRED for ping (≤800). For request_meeting: becomes context_summary. For send_intro: becomes intro_pitch (≤4000). Sanitized server-side.
intentYesping = heads-up, no reply expected, ≤800 chars, no URLs (anti-phish). request_meeting = meeting invite; body becomes context_summary (recipient counter-proposes windows in v1). send_intro = broker an introduction TO the recipient; requires intro_target + intro_pitch (=body) + expected_outcome + consent_target_has_opted_in (anti-spam: you MUST have the target's consent). message = open a back-and-forth conversation; body is the message (≤4000, URLs allowed — it's dialogue with a policy-gated member, not a cold ping); the recipient replies via clarify and either side closes via decline/withdraw.
idempotency_keyNoOptional deterministic key. Same (token, key) within 24h returns the same response. Use UUIDs or a deterministic-from-source hash.
expected_outcomeNosend_intro: what you're asking the recipient to do. REQUIRED for send_intro.
intro_target_orgNosend_intro: the intro target's org (optional).
intro_target_nameNosend_intro: name of the person being introduced TO the recipient.
intro_target_contextNosend_intro: short context on who the target is / why (optional).
intro_target_linkedin_urlNosend_intro: the intro target's LinkedIn URL (optional).
consent_target_has_opted_inNosend_intro: you attest the intro target has consented. MUST be true — the server rejects false/absent (anti-spam, DESIGN §5).

TDQS

A4.7/5.0
Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

Annotations are all false (no read-only, idempotent, etc.), so the description carries the full burden. It discloses server-side sanitization, persistence, policy evaluation (store+notify, queue-for-tap, silent drop, refuse), idempotency caching, the exact return shape, the mcpError on closed inbox, and the opaque silent-block behavior. This is exceptionally transparent for an agent, leaving no hidden side effects.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is long but well-ordered: purpose, server behavior, scopes, send_intro special requirements, idempotency, return shape, precheck, and a v1 note. Each sentence carries information with minimal fluff. It is not overly verbose given the complexity of 10 parameters and 4 intents, but a future revision could condense the server-side details slightly without loss. Structured and front-loaded with the core action.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

Given 10 parameters, multiple intents, no output schema, and no annotations to lean on, the description is remarkably complete. It explains the return shape, edge cases (blocklist, closed inbox), prerequisites (directory lookup), and per-intent requirements. An agent has everything needed to call this tool correctly. No gaps remain for correct invocation.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters5/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Though schema coverage is 100%, the description adds substantial meaning: it defines body per intent (≤800 for ping, becomes context_summary for request_meeting, intro_pitch for send_intro), states that send_intro requires intro_target_name + body + expected_outcome + consent_target_has_opted_in=true, and explains idempotency_key's 24h caching. It also clarifies the 'message' intent not mentioned in the main description. This goes far beyond the schema.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description states a specific verb ('Route a typed intent') and a clear resource ('another IC member's agent inbox'), enumerating the three valid intent types (ping / request_meeting / send_intro). It differentiates itself from sibling tools like ic_agent_inbox_reply by focusing on sending, not responding. The intent and scope are unmistakable.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines4/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

The description provides an explicit PRECHECK ('call ic_agent_directory_lookup first') and explains when the tool will refuse (closed inbox). It also warns that send_intro requires extra params and consent. However, it does not explicitly contrast with sibling tools like ic_agent_inbox_reply (though that is semantically different), leaving the agent to infer when to use this versus a reply. The precheck and per-intent requirements are strong, but an explicit 'when-not-to-use-except-reply' would be clearer.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

Try in Browser

Glama MCP Gateway

Add one secure layer between your agents and this server.

TDQS

A3.7/5.0
Disambiguation4/5

Most tools are clearly scoped to distinct actions (e.g., ic_hack_apply vs. ic_hack_register, ic_rooms_create vs. ic_rooms_join). A few pairs could confuse an agent: floor10_submit_highlight vs. floorcast_push both submit HighlightStories but to different queues, and ic_directory_search / ic_agent_directory_lookup / ic_admin_list_members overlap in searching members. Overall, the long descriptions help disambiguate, but the volume requires careful reading.

Naming Consistency3/5

The dominant pattern is ic_<domain>_<verb>_<object> (e.g., ic_admin_list_pending_events, ic_headsets_checkout), but there are notable deviations: floor10_* and floorcast_* prefixes break the ic_ convention, and a few tools use noun-style names (ic_health, ic_capabilities, ic_donations_total). Verb placement also varies (get_* vs *_get, e.g., ic_get_my_membership vs. ic_membership_set_profile). Still, most names are readable and predictable.

Tool Count1/5

175 tools is an extreme count for a single MCP server, far beyond the 50+ threshold that indicates an unwieldy surface. While the platform covers many domains (events, files, hackathon, headsets, prints, rooms, etc.), bundling everything into one server makes discovery and selection difficult. This would be better split into several narrowly-scoped servers.

Completeness4/5

The tool set covers nearly every lifecycle for each domain: CRUD for files/folders, full hackathon admissions and judging, headset lending with waivers and incidents, print farm submission and handoffs, and room coordination. Minor gaps exist: no delete for files/folders, no cancel for events, and some actions (like revoking a Z.ai key or tearing down a room) are explicitly left to human console use. Overall, the surface is remarkably comprehensive for the stated scope.