[AISP-1533] Signals MCP release

[Peter Perlepes (igneel64)]

What changed?

Signals MCP server first release with Agentic Context fetching.

[github-actions]

Missing SEO metadata

The following markdown files are missing required metadata fields:

• release-notes/signals-mcp-server-for-agents/index.md: missing fields: keywords, sidebar_label

Required fields

The file metadata is important for SEO and marketing. All markdown files, except for those with filenames starting with _, should include:

• title: Full, descriptive page title
• sidebar_label: Short title for navigation sidebar (can be the same as the main title)
• description: One to two sentences summarizing the page contents
• keywords: Array of marketing/SEO keywords

Please add the missing metadata.

[cloudflare-workers-and-pages]

Deploying with    Cloudflare Workers

The latest updates on your project. Learn more about integrating Git with Workers.

Status	Name	Latest Commit	Preview URL	Updated (UTC)
✅ Deployment successful! View logs	documentation	6415f81	Commit Preview URL Branch Preview URL	Sep 03 2026, 09:39 AM

[Peter Perlepes (igneel64)]

Claude (@claude) review again

[claude]

Writing quality review

Solid page overall — the tab structure, admonitions, mdx-code-block imports, internal link format (/index.md), and frontmatter all follow the repo conventions. Suggestions below, ordered roughly by impact.

docs/signals/applications/index.md

"the same data" is inaccurate. The MCP server serves agentic contexts only, not attributes or interventions, so telling readers the agent can "read the same data" through it overstates parity with the SDKs. "read ... as tools" also reads awkwardly. Suggested rewrite:

If the consumer is an agent, it can discover and read your agentic contexts as tools through the Signals MCP server.

Bullet phrasing. The other bullets in the list start with a verb describing what the reader does ("fetch calculated attribute values", "receive intervention payloads"). Consider matching: "discover published agentic contexts from an agent and read the ones that fit the task".

docs/signals/applications/mcp-server/index.md

Missing link to the page where agentic contexts are defined. The page repeatedly tells readers that "what you write when you define a context" steers the agent, but every agentic-contexts link points at retrieve agentic contexts. Add a link to /docs/signals/agentic-contexts/index.md (where the description and prompt fields are documented) the first time you mention defining a context.

State the server URL in prose. The release note says it plainly — "served at /mcp on your Signals API URL" — but the docs page only shows it inside code samples and the Inspector steps. Worth adding one sentence at the top of "Connect to the server", with a link to connection credentials for the Signals API URL.

Credential naming. /docs/signals/connection/index.md calls the values "API key" and "API key ID". The page says "your Snowplow API key and key ID" — use "API key ID" for consistency.

Passive voice and repetition in "Inspect the server". "The server is fully behind authentication, so without both headers the connection is rejected during initialize" is passive and repeats "The server rejects connections that don't send both headers" from the Connect section. Consider trimming to something like: "The server rejects the connection during initialize if either header is missing."

Prompt example uses JS template syntax. In a text block, ${domain_sessionid} reads as a literal string rather than a placeholder:

The current user's Signals session identifier is: ${domain_sessionid}.

Use a plain placeholder (<domain_sessionid>) or state that you interpolate it. The wording also differs from the Vercel example above it ("The current Signals session identifier is:") — make the two match.

Dense paragraph in "Available tools". The paragraph after the table packs four separate ideas into one block: the two-call pattern, published-only listing, the unknown-name behavior, and the prompt. Split it into two or three sentences grouped by idea, and consider moving the prompt point next to the get_agentic_context row.

Repeated phrasing. "whichever ones the task warrants" / "whichever one fits the task" / "the ones that fit the task" appears four times across the three files. Vary or cut.

Smaller wording items:

• "the standard way an agent discovers and calls tools" — "an open standard for how an agent discovers and calls tools" is more accurate and less absolute
• "which is separate from ai and doesn't pull it in, so install both" — convoluted; "ai doesn't include @ai-sdk/mcp, so install both alongside your model provider" is clearer
• "so you don't write retrieval code for each agentic context" → "so you don't have to write retrieval code"
• "so you have to tell your agent" → "so you need to tell your agent"
• "ignored whenever transportOptions is set" → "when"
• "the agentic contexts your deployment has published" → "the agentic contexts you've published" (the style guide asks for direct "you" address; the deployment isn't the actor)
• const domain_sessionid is snake_case in a TypeScript sample — consider domainSessionId
• sidebar_position: 27 sits between the neighboring 25 and 30, so ordering is right, but siblings otherwise use 10/20/25/30

release-notes/signals-mcp-server-for-agents/index.md

"now" is on the style guide's banned list ("Stay in the present tense. Never use 'currently', 'now'..."). It appears in both the description and the first line. "Signals deployments expose their agentic contexts over..." carries the same meaning.

Missing comma before "and" joining two independent clauses: "You publish agentic contexts from Console or the Python SDK**,** and your agent can pick them up automatically."

Passive / weak phrasing in the disambiguation paragraph. "This is not to be confused with" and "is there to" are both indirect. Suggested:

Don't confuse this with the Snowplow MCP server, which connects an assistant such as Claude Code or Cursor to your Snowplow account for managing resources like data structures, pipelines, and Signals configuration. The Signals MCP server serves Signals resources, such as agentic contexts, to an agent running in your own product.

Wrong link target for publishing. "You publish agentic contexts from Console or the Python SDK" — the publishing workflow is documented at /docs/signals/agentic-contexts/, not the retrieve page linked in the paragraph above. Link that one here.

Dates. The release note is dated 2026-09-01 and the docs page 2026-09-03. Worth aligning both to the publish date.

[jborlase-snowplow]

LGTM!

Few small non-blocking recommendations:

• We should add a note about the MCP server to the Use in your applications > Agentic Context page.
• We should add it to the "Connect to Signals" page somewhere as we also use that to discuss SDK usage and retrieval.