We use cookies to enhance your browsing experience and analyse our traffic. By clicking “Accept All”, you consent to our use of cookies according to our Cookie Policy. You can change your mind any time by visiting out cookie policy.
TL;DR: Terra API now ships ten agent skills and a docs MCP server, so Claude, Cursor, Copilot and any agent that reads the Agent Skills standard can build your integration from the current API instead of from what the model remembers. Install them with one command, npx skills add tryterra/agent-skills, or let the Terra CLI set up every agent on your machine with terra agent setup. The skills carry the rules that matter most across real integrations: verify webhook signatures on the raw body, acknowledge inside the timeout, dedupe chunked deliveries, and never overwrite a score with a null.
Coding agents are good at the parts of an integration they have seen thousands of times. They are much less reliable on an API with a long list of endpoints, a large number of health data sources, and a data model that appears rarely in training data. Terra API is exactly that kind of surface. The agent skills give the agent the current rules instead of leaving it to guess.
Ten skills, with the webhook and data rules first
The tryterra/agent-skills repository ships ten skills, each covering one part of the product:
Managing the account over the admin MCP server from chat assistants, with dashboard approval for sensitive changes
The one most integrations start with is terra-unified-api. It carries 21 rules in five categories, ranked by impact, so an agent reads the webhook and data rules first. Four examples, each linked to the rule itself:
Verify the signature before parsing. Compare the terra-signature HMAC against the raw request body. Parsing the JSON first changes the bytes, and the check stops meaning anything.
Acknowledge inside the timeout. Return a 200 within the eight-second default window and do the real work asynchronously. A slow handler gets retried, and a retry looks exactly like new data.
Dedupe deliveries. Key on X-Terra-Trace-Id, because large payloads arrive in several chunks that share one reference.
Never overwrite a score with a null. Standard fields carry a superset guarantee and enrichment scores do not, so COALESCE the new value with the stored one and a user never loses a number they have already seen.
The rules are written as explicit instructions rather than descriptions, because that is what changes agent behaviour. "Only overwrite if the timestamp is newer" produces better code than a paragraph explaining why timestamps matter. Every rule comes from a pattern that shows up across real integrations, and every rule file says what to do and why.
Get the latest Terra Research reports and insights every week as soon as they're published.
The MCP servers keep the agent on the current docs
Skills carry judgment about how to use the API. A model's memory of field names, endpoints and schemas is frozen at its training cutoff, and the live specification is not, so the agent also needs a way to read the current docs while it works. Two MCP servers do that. The docs MCP server at https://docs.tryterra.co/~gitbook/mcp answers lookups against the current documentation, and the dashboard setup modal installs it for your editor as terra-docs. The Terra API admin MCP server goes further: it lets the agent manage your account, with approval in the dashboard for sensitive changes, and it answers documentation questions through its docs_ask tool. Either way the agent is reading the API as it is today, not as the model remembers it.
Skills load only when a task matches, so the agent spends context on Terra API knowledge at the moment it needs it. A quiet first session is normal.
The recommended path is the Terra CLI, which detects the agents on your machine and writes the skills where each one reads them. Install the CLI first, then run the setup:
curl -fsSL https://cli.tryterra.co/install.sh | sh
terra agent setup
Homebrew (brew install tryterra/tap/terra) and npm (npm install -g @tryterra/cli) work too. Run terra agent setup --status-only to see what is installed without writing anything. Once the skills are in place, a new session only needs the prompt "Get started with Terra API": the agent checks that the CLI is signed in, finds your environments, and reports which providers are on and where webhooks go.
Or install every skill directly with skills.sh:
npx skills add tryterra/agent-skills
The same set ships as a plugin for Claude Code, Codex and Cursor. The plugin also connects your agent to your Terra API account through the CLI, so it can configure webhooks, providers and keys without you clicking through the dashboard, and it asks before any sensitive change. In Claude Code:
In Codex, run codex plugin marketplace add tryterra/agent-skills and pick Terra API in the plugins browser. In Cursor and other Agent Plugin hosts, run npx plugins add tryterra/agent-skills. If the CLI is missing or logged out, the plugin offers to install it and log you in, and it never does so without asking. The plugin updates on its own; skills installed with npx skills add do not, so run npx skills update -y to pull the latest versions.
The dashboard has a setup modal in the navigation bar with a copy-paste command or an install link for Claude Code, Codex, VS Code and Cursor, including the MCP configuration. All of the skills are public on GitHub under an MIT licence, so you can read every rule before your agent touches your code, and open an issue when one disagrees with what you see in production.
Try it on your next feature
The quickest test is your next feature: install the skills, point your agent at it, and read the diff it produces. The setup guide and the per-agent instructions are at docs.tryterra.co/developer-tools/terra-cli/coding-agents.
