Blog · July 9, 2026 · 3 min read
Connect any agent to DigestOps over MCP in three minutes
Claude Code, Codex, Cursor, VS Code, or any MCP client: one command against the public read server, four read-only tools, and a free key when you need volume.
DigestOps exposes its entire verified library — living answers, corroborated claims, scored evidence, and change history — through one public MCP server. Any MCP-capable agent can read it: Claude Code, Codex, Cursor, VS Code, or your own client. No account is required to try it, and the whole setup is genuinely one command.
What your agent gets
The public read server at https://digestops.com/api/mcp speaks Streamable HTTP and advertises exactly four read-only tools:
Every tool is annotated read-only and idempotent, so agent runtimes can call them freely without confirmation prompts. Answers come back with citations, corroboration counts, and a freshness envelope — your agent knows not just what is true, but how well-supported and how recent it is.
One command per client
Claude Code
Verify with claude mcp list, then ask Claude to call the DigestOps catalog tool.
Codex
Or declare it in ~/.codex/config.toml:
Verify with codex mcp list, then ask Codex to list the available DigestOps topics.
VS Code
Open “MCP: List Servers” from the Command Palette and start DigestOps.
Cursor and other clients
Any client that reads an mcpServers block works with one entry:
For Cursor, drop it in .cursor/mcp.json (or use the one-click install on the ); other clients take the same JSON in their MCP configuration. Restart the client and confirm that catalog, read, find, and diff are visible.
Three calls to try first
Once connected, these prompts exercise the whole surface:
- Morning brief — “What changed across DigestOps topics in the last 24 hours?” The agent calls
read(view=changes, window=24h)and returns only deltas, not a feed to scroll. - Deep dive — “What are the current best practices for coding agents? Use DigestOps.” That's
read(question=..., view=answer)followed byfind(query=...)for supporting evidence. - Fact check — “Is this claim corroborated?”
find(query="...", min_corroboration=2)returns only claims that independent authors agree on.
Browse the live catalog on the — everything you can see there, your agent can query.
The call that matters: diff
Feeds tell you what just happened; models are frozen at their training cutoff. The question agents actually need answered is what changed since I last checked — and that is a first-class call:
The response is a structured changelog: claims that were added, strengthened, weakened, or overturned since that date, each with citations. It turns a recurring research task into a single idempotent tool call.
When to add a key
Anonymous calls get a taste: the catalog plus answer verdicts. A free key — created in after signup, shown once, revocable any time — unlocks the last seven days at 100 requests per day. unlocks full history, as_of answer replay, diff, series, and production volume.
Adding the key is the same one-liner with an authorization header:
FAQ
Does the public MCP server let agents write anything?
No. The public server is strictly read-only — four tools, all annotated idempotent and non-destructive. Refresh requests live on a separate OAuth-protected action server, so a misbehaving agent can never mutate the library through the read path.
What happens when I hit the free-tier limit?
Requests over the daily allowance return a clear rate-limit error with a reset time; nothing breaks mid-session. If your agent needs sustained volume or historical replay, that is what the Pro plan is metered for.
Which MCP transport does DigestOps use?
Streamable HTTP, the current MCP standard. Any client that supports HTTP transport works out of the box; no SSE fallback or local proxy is needed.
Connect once, then let your agent ask the recurring questions — the covers authentication, views, and the REST fallback in depth.