Docs · MCP

Plug DigestOps into your agent

Four public read tools over Streamable HTTP. Every response carries a freshness envelope (data_as_of, staleness) so your agent always knows how current the data is.

1 · Connect

Connect your agent

Install the public read server

No account is required to test it. Add a free key when you need higher limits and recent evidence.

Streamable HTTP

Run in a terminal

shell

claude mcp add --transport http digestops https://digestops.com/api/mcp

VerifyRun `claude mcp list`, then ask Claude to call the DigestOps catalog tool.

Need authenticated or unattended access?

API keys are shown once and should not be embedded in install links. OAuth-capable action clients authenticate in the browser.

Create free account

Anonymous calls get a taste (catalog + answer verdicts); a free key unlocks the last 7 days at 100 req/day; Pro unlocks full history, as_of, diff and series.

2 · Keys and protected actions

Use an API key for unattended jobs and clients that cannot complete OAuth. Keys are created from a free account, shown once, and can be rotated or revoked at any time. Do not put a key in a shared repository or an install-link URL.

The protected action server lives at https://digestops.com/api/actions/mcp. It uses DigestOps OAuth and browser sign-in, so compatible clients do not need a copied secret.

3 · Public read tools

tools

catalog(theme?)                    what exists: themes, questions, frames, views
read(question|theme|entity, view)  view = answer | top | changes | series | profile
find(query, theme?, min_corroboration?)   search evidence & claims
diff(question, since)              everything that changed since a date  [Pro]

every read tool also takes format: "json" (default) | "markdown"

The write action refresh(theme) lives at https://digestops.com/api/actions/mcp. It requires DigestOps OAuth, the refresh:write scope, the correct resource audience, and an active Pro entitlement.

The envelope on every response: freshness (data_as_of, staleness_seconds, refresh_state, next_scheduled), provenance (attribution on the free tier) and quota (used/limit/resets_at — daily on free, monthly fair-use on Pro).

4 · The call that matters

Your model knows the world up to its training cutoff. One call closes the gap:

call

diff(question="coding-agent-best-practices", since="2026-01-01")
→ current answer + version history + new claims + what got corroborated since

5 · Typical agent recipes

recipes

morning brief   read(view=changes, window=24h)
deep dive       read(question=..., view=answer) then find(query=...)
tool choice     read(question=ai-coding-tool-rankings, view=top)
fact check      find(query="...", min_corroboration=2)
history replay  read(question=..., view=answer, as_of="2026-06-01")  [Pro]

Free-tier responses require attribution: link the answer page when you use its content. REST access (legacy v1) is documented at /docs.