distribution.tech MCP¶

Every founder who has bought a post has bought it the same way: a DM, a screenshot of a rate card, a PayPal transfer, and no way to prove the post ran. This connector is that market with a price, an approval step and evidence attached, exposed to an agent as tools.

Server type: Remote (Streamable HTTP)
Endpoint:    https://app.distribution.tech/api/mcp/guest   (no key)
             https://app.distribution.tech/api/mcp         (API key or OAuth)
Auth:        guest tier keyless; buying needs an approved brand and a key
Tools:       6 on the guest surface, 22 documented on the account surface
Limits:      60 requests per 10 minutes per client address (guest)
Pricing:     one all-in price per post, one total per plan

What it does¶

The guest surface is the first half of the deal and it is readable without an account:

inventory answers how much of the network is on sale today per channel, as counts only, never names: verified, listed, bookable, and priced units.
resolve_audience takes a described profile (product, audience, geography, at least one of product or audience) and returns the segments it matches, the regions it resolves to, the coverage per channel and whether that coverage is thin.
plan_quote_guest estimates the all-in price range per post in one format (post, quote or thread) across every creator on sale now, and multiplies by the number of posts named (1 to 50).
guest_status reports what the connection can do without a key, the guest limits, and the claim path.
brain_get returns the saved company context, which as a guest is version 0 and empty.
feedback_send is stated to need an account and answers with a claim rather than sending.

The account surface, documented at /docs/agents, is 22 tools that carry a campaign from conversation to evidence: creators_search, creator_get, creator_ask_price, plan_quote, create_campaign, plan_add_creator, plan_remove_creator, plan_move_dates, plan_swap_creator, set_creator_guidelines, approve_plan, archive_campaign, rename_campaign, delete_campaign, campaign_status, evidence, results, brain_get, inventory, resolve_audience, feedback_send and feedback_list.

What this pass measured¶

initialize on the guest endpoint returned HTTP 200 in 0.34 seconds, protocolVersion 2025-06-18, serverInfo name distribution at version 1.37.0, with tools.listChanged, resources and prompts all advertised. tools/list returned 5,393 bytes and all six guest tools with full schemas.

guest_status then described its own limits precisely: signed_in: false, access: "guest", answers_now listing five tools, needs_account listing feedback_send, a tool-call limit of 60 per 10 minutes per client address, and a separate limit of 3 plan_quote_guest calls per 7 days per client address. That second limit is the important number for anyone planning to use this as a price research tool: three quotes a week.

inventory, computed at 2026-11-01T04:02:35Z with coverage as of 04:02:33Z, returned one channel, x_creators, sale: "live", level: "strong", with 1,005 verified nodes, 1,005 listed, 86 bookable now and 255 posts carrying a current price.

resolve_audience on a described profile (a B2B AI data layer for ecommerce and SaaS operators, US) resolved to three segments, ai (AI builders and early adopters), saas_b2b (B2B software buyers and operators) and finance (Finance and investors), region us, unresolved: false, and then narrowed the inventory to 792 approved creators, 63 bookable and 187 priced posts, with thin: false. The gap between 1,005 and 792 is the whole argument for the tool: an audience filter removed 213 creators who exist but do not reach that audience.

plan_quote_guest failed. The call returned isError: true with the text unavailable: The network could not be read just now. Try again. No price was obtained, one of the three weekly attempts was spent, and the defect is recorded below rather than papered over.

The containment worth reading twice¶

The spending model is the reason this entry is worth a guide rather than a line in a list. Per the agent documentation, a key carries a mandate that caps what it can spend, signing a plan produces a payment link inside that mandate, and a brand cannot buy at all until it has been reviewed and approved. So the failure mode an operator fears, an agent committing a budget because it was told to "get us some reach", is bounded by a number a human set before the agent ran.

The second containment is the guest tier's own design: counts, not names. A guest can learn how deep the network is and how well it matches an audience without reading a single creator's identity, and creator names arrive on a plan before signing, with two alternates and the ability to swap a placement.

The operator decision this answers¶

Paid distribution is the leg most product teams cannot plan, because the unit price is opaque until someone answers an email. Three numbers decide it, and this surface returns all three before a human talks to anyone: how many creators reach the audience (792), how many are purchasable today (63), and how many posts are actually priced (187). A thin market is visible as a thin: true and a low bookable count rather than discovered after a week of DMs.

The workflow is the interesting part for an agent-first operator: the agent reads coverage for a described audience, asks the price in the format it wants, assembles a plan with named creators and alternates, and stops at an approval. That is a procurement process a founder can audit, which is more than can be said for a DM thread and a bank transfer.

Where it sits in this catalogue¶

Nothing here does what this does. The nearest entries touch one edge of it:

B2B Creators MCP is a creator and expert sourcing surface, which answers who exists, not what a post costs or whether it ran.
PR CY MCP is press and media placement rather than creator placement.
The social scheduling and publishing entries (Blog2Social, the bulk publishers) post content you already own to accounts you already control, which is the opposite side of the market: owned distribution versus paid distribution.
The ad connectors buy impressions inside a platform's auction with a self-serve interface and no human seller, so there is no approval or proof step to model.

What this pass did not establish¶

The price is unmeasured, and it is the number that matters. plan_quote_guest returned an isError payload on the first attempt, and two of the three weekly guest quotes remain, so the all-in range per post is an unknown rather than a finding. Everything about the buying half is documentation: no brand approval was sought, no key was created, so create_campaign, approve_plan, evidence, results and the mandate enforcement are read from /docs/agents and not driven.

Also unread or unverified: what share of a placement's price reaches the creator; whether evidence is a screenshot, a link, a platform metric or all three; whether the 3-campaigns-a-day limit applies per brand or per key; and whether the 1,005-creator inventory is audited for follower authenticity, which the site's "reviewed creators" language implies and this pass did not test.

FAQ¶

Does it need an API key?¶

Not to read the market. The guest endpoint is keyless and returns coverage, audience fit, price estimates and status. Buying a placement needs an approved brand plus an API key from the workspace, or OAuth.

Can an agent spend without approval?¶

It can sign a plan, and signing produces a payment link rather than a charge, inside a mandate the brand set on the key. Publishing a campaign to the network is gated on brand review, and the documented flow puts approve_plan after a plan exists, with evidence and results after it runs.

Why did the price quote fail?¶

plan_quote_guest returned unavailable: The network could not be read just now. Try again. That is a transient-service error at the vendor, not a permissions error, and it cost one of three weekly guest quote attempts. Retry later in the week rather than in the same minute.

How current is the inventory?¶

The response stamps both computed_at and coverage_as_of. In this call they were 2.4 seconds apart, so the counts are a live read rather than a nightly snapshot, and a bookable count that moves between two calls minutes apart is a marketplace, not a bug.

Is this the same as buying ads?¶

No. Ads are an auction with self-serve targeting and no seller. This is a marketplace of named humans whose posts are priced individually, with approval, mandate and per-placement proof.

Integration with CorpusIQ¶

CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.

This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.

See what CorpusIQ connects