Skip to main content
Glama

create_campaign

Destructive

Create a LinkedIn outreach campaign from a natural language description.

Instructions

Create a LinkedIn outreach campaign from a natural language description.

Finds people on LinkedIn who match the description and saves a draft;
nothing is sent until campaign(action='launch'). Use it for sales
prospecting, recruiting and candidate sourcing, research and
user-interview recruitment, job-search networking, investor and partner
outreach, vendor scouting and event invitations.

Reaching people you can already name — an article author, a warm intro, a
speaker, one founder a customer mentioned — is the `people` argument: pass
their profile URLs and the campaign is seeded from exactly them, with no
LinkedIn search. Everything else is unchanged: draft until launch, rate
limits, sending window, opt-outs.

Describe who you need to reach and HeyLead finds them on LinkedIn:
customers (lead generation, B2B prospecting, cold outreach), candidates
(recruiting and sourcing), hiring managers and referrals (job search),
investors and partners, vendors, and user-interview or research participants.
On first campaign, project_brief is asked explicitly, in the goal's words
(see project_brief below) — a homepage alone is not enough.

Args:
    target_description: Who to target (e.g., "CTOs at fintech startups",
        "freelance UX designers in London", "yoga studio owners in California")
    campaign_name: Optional name for the campaign.
    icp_id: Optional ID of a saved ICP from generate_icp. If provided,
        uses the saved ICP's enriched LinkedIn codes for precise targeting
        instead of generating a new one.
    company_context: Optional. Your website URL or 1-2 sentences about your
        product/company. Copied into project_brief when project_brief is omitted.
    project_brief: Optional. Full project paste the model sees, in the goal's
        words. sell: what you offer and who it is for. job_search: the role,
        the kind of company, what you bring. hire: the role, who fits it, what
        it offers. partner: what you want from a partner, what you bring.
        buy: what you need, by when and how much, what a vendor must confirm.
        research: what you research, who you want to hear from, what you ask.
        Required before launch, resume, or auto-send.
    mode: Always autopilot (copilot mode was removed). Whether opening
        DMs, follow-ups and answers to replies wait for a person is the WORKSPACE's approval mode,
        not this: a hosted workspace that never chose sends them without
        review; scheduler(action='approval_mode', mode='require_approval')
        holds them (inspect(action='waiting') lists what waits).
    company_url: Optional LinkedIn company URL for account-based targeting.
        Searches for employees at that specific company matching the ICP.
        Example: "https://www.linkedin.com/company/google"
    voice_mode: "text_only". Messages are text.
    connections_only: "on" to create a DM-only campaign targeting existing
        LinkedIn connections. Skips invitations and warm-up — sends DMs
        directly to people you're already connected with. Use when user says
        "existing connections", "DM my network", "message my connections".
    exclude_connections: ON BY DEFAULT for a new campaign: nobody who was
        already a 1st-degree connection before this campaign started is
        reached — refused at enrolment and skipped at send time rather than
        DMed. People who accept this campaign's own invitation still get
        the opener. Pass "off" to include existing connections (or turn it
        off later in the campaign settings). Defaults off only for a
        connections_only campaign. Use "off" when the user says
        "include my existing connections"; the old "on" is still accepted
        for "don't message my existing connections", "cold only",
        "skip people I already know". Cannot be combined with
        connections_only, which is its exact inverse.
    people: LinkedIn profile URLs or public identifiers, comma or newline
        separated (e.g. "linkedin.com/in/jane-doe, linkedin.com/in/john-doe").
        The campaign is seeded from exactly these people: no LinkedIn
        search runs, the goal <-> ICP audit is skipped (the audience is
        stated, not inferred), a low ICP score does not drop anybody, and
        discovery stays off so nothing tops the queue up with strangers.
        Use it whenever the user names who to reach. Cannot be combined
        with connections_only.
    campaign_type: Prompt family: "outbound" (default) or "job_search".
        job_search writes a job-search campaign: the invitation note and
        the first DM may name the recipient's company and the role, use
        one credible proof point at most and never list a CV. InMail is
        not routed by this switch. Pair with connections_only="on" to
        write to people the sender is already connected to.
    force: True to create the campaign even when the goal <-> ICP audit
        returns `mismatch` (the ICP holds no plausible buyer for the goal).
        Leave False; a `partial` verdict never blocks, it only warns.
    goal: What the campaign is for: "sell" (default), "job_search",
        "hire", "partner", "buy" or "research". It sets campaign_type and
        campaign_intent, picks whose profile the ICP describes, and picks
        the fit question the goal <-> ICP audit asks. hire, partner and
        research run on your project_brief until their message sets exist.
    message_language: The language every message to a prospect is
        written in: "en", "prospect" (their latest message, else their
        LinkedIn profile's language, else English) or a code: uk, de, fr,
        es, pl, pt, it. Never set this without asking the user. Empty
        leaves it unset: messages go out in English, and launch asks once
        when the brief is in another language.

Tool call template

Replace each placeholder with a value matching the input schema. Only required parameters are included; add optional parameters as needed.

{
  "id": 1,
  "jsonrpc": "2.0",
  "method": "tools/call",
  "params": {
    "arguments": {
      "target_description": "<target_description>"
    },
    "name": "create_campaign"
  }
}

Input Schema

NameRequiredDescriptionDefault
goalNo
modeNoautopilot
forceNo
icp_idNo
peopleNo
voice_modeNotext_only
company_urlNo
campaign_nameNo
campaign_typeNo
project_briefNo
company_contextNo
connections_onlyNo
message_languageNo
target_descriptionYes
exclude_connectionsNo

Output Schema

NameRequiredDescriptionDefault
resultYes

Schema Changelog

Changes observed during successful MCP inspections.

  1. Changed1 schema field changedv0.10.454
    • addedInput schema / properties / message_language
      Added value: +{
      +  "default": "",
      +  "title": "Message Language",
      +  "type": "string"
      +}
  2. Changed1 schema field changedv0.10.398
    • addedInput schema / properties / goal
      Added value: +{
      +  "default": "",
      +  "title": "Goal",
      +  "type": "string"
      +}
  3. Changed1 schema field changedv0.10.389
    • addedInput schema / properties / people
      Added value: +{
      +  "default": "",
      +  "title": "People",
      +  "type": "string"
      +}
  4. First observedv0.10.375

TDQS

A4.8/5.0
Behavior5/5

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

Goes well past the annotations: it discloses that nothing sends until launch, that `exclude_connections` is ON by default and refuses people at enrolment and send time, that `force` only overrides a `mismatch` audit verdict while `partial` merely warns, that `people` disables discovery and skips the ICP audit, and that approval behavior lives in the workspace's approval mode rather than this tool. This is meaningful context for a destructive, open-world write that annotations only label generically.

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

Conciseness3/5

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

Front-loaded and well organized into a lead paragraph plus per-argument notes, but the audience list ('sales prospecting, recruiting and candidate sourcing, research...') is restated almost verbatim in the following paragraph about target_description, which is pure duplication. Several argument notes are discursive rather than tight.

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?

For a 15-parameter mutating tool with an output schema and only generic annotations, the description supplies everything an agent needs: creation vs send boundary, defaults, mutually exclusive flag pairs, and failure/override semantics. No material gap remains, and return values are covered by the output schema.

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?

Schema description coverage is 0% and the description compensates fully, documenting all 15 parameters with defaults, examples and interaction rules (e.g. company_url format, goal values and their downstream effects, message_language codes and the 'prospect' fallback chain, connections_only vs exclude_connections conflict). Little is left for an agent to infer from bare field names.

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?

States a concrete verb and resource ('Create a LinkedIn outreach campaign from a natural language description') and immediately scopes the outcome: it finds matching people and saves a draft rather than sending. It also separates itself from the sibling `campaign` tool by naming `campaign(action='launch')` as the thing that actually sends.

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

Usage Guidelines5/5

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

Gives explicit routing rules: use `people` when the user already names who to reach (no LinkedIn search), use `target_description` otherwise, and use `connections_only` only for existing-network DMs, which is stated as the exact inverse of `exclude_connections` and therefore mutually exclusive. It also names exclusions and prerequisites (project_brief required before launch/resume/auto-send, 'Never set message_language without asking the user').

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