Tools
A signed-in connection gets nineteen tools. Your role decides what each one can see and change: the same projects, environments, variables, secret files, shared accounts and documentation you can use in the dashboard. Every value-returning call and every write is audited against you.
Read tools#
| Tool | What it does |
|---|---|
envpilot_list_organizations | The organizations you belong to, your role, and whether MCP is on their plan |
envpilot_list_projects | List the projects you can see |
envpilot_get_variables | Variables for a project + environment, with optional keys / prefix / metadata_only |
envpilot_get_variable | One variable by exact key |
envpilot_search | Search project names/slugs and variable keys, never values. Bounded to 20 projects / 100 matches |
envpilot_search_docs | Full-text search over published documentation: titles and page bodies. Never returns drafts |
envpilot_get_doc | One published page, or a draft you wrote: its outline, one section, or up to 60,000 characters at a time |
envpilot_list_files | Secret files in a project: path, size, sha256, environments. Never contents |
envpilot_get_file | One secret file's contents by path. Text as utf8, binary as base64 |
envpilot_list_accounts | Shared accounts in a project: name, website, environments. Never credentials |
envpilot_get_account | One shared account's username and password by exact name |
envpilot_list_requests | Change requests you filed and their status: pending, applied, rejected (with the reason), canceled, expired |
Write tools#
An agent can add and change the same things you can in the dashboard, under your role:
| Tool | What it does |
|---|---|
envpilot_create_variable | Create a variable with a value in one or more environments |
envpilot_update_variable | Change the value or description of the variable that holds a key in one environment |
envpilot_upload_file | Store a secret file at a path, or replace its contents with replace: true |
envpilot_create_account | Store a shared account's login |
Values, file contents and credentials are encrypted before they are stored, exactly as in the dashboard. In Claude every write tool carries destructiveHint, so Claude asks before each one.
Protected environments become requests#
A write into a protected environment is never applied over MCP. The tool files a change request instead and answers status: "proposed" with a requestId and a reviewUrl. A second person approves it in the dashboard; until then nothing changes. envpilot_list_requests shows whether it was applied or rejected. There is no break-glass override over MCP.
Updating a variable changes every environment it covers#
A variable can cover several environments with one value. envpilot_update_variable finds the variable that holds the key in the environment you name, and the response lists every environment it covers, because they all change together.
Files#
envpilot_upload_file refuses .env and .env.* paths: store those values as variables. A path can exist once per environment; when the same path is stored for several environments, pass environment to envpilot_get_file. File size and count limits are your plan's, the same as the dashboard's.
Draft tools#
An agent can write documentation drafts. Every draft ends at a person:
| Tool | What it does |
|---|---|
envpilot_create_doc | Propose a documentation page. Always creates a draft; a person publishes it |
envpilot_list_doc_drafts | The drafts you wrote in a project, newest first |
envpilot_update_doc_draft | Replace, append to, or replace one section of a draft you wrote |
A draft page is invisible to your team, and to every other agent, until a person publishes it. You can read your own drafts back with envpilot_get_doc, including the outline whose section ids replace_section needs. In Claude the two write tools carry destructiveHint, so Claude asks before each one.
Variable requests were only ever available to API keys and were removed with them.
Documentation tools in detail#
Drafts are invisible#
envpilot_create_doc always writes a draft. Nothing over MCP can
publish. envpilot_search_docs never returns drafts, and envpilot_get_doc
returns a draft only to the person who wrote it — anyone else gets "not
found", even with the id. That gate is what stops a page one agent wrote from
reaching another agent's context before a person has read it.
Page count is capped by plan#
envpilot_create_doc fails once the project or the organization is at its
page limit — 10 per project and 25 per organization on Free, unlimited on Pro.
The caps are the same numbers the dashboard enforces, so an agent cannot use
the MCP surface to walk around them. See Plans & Limits.
Long pages: write in parts, read in pages#
A single tool call carries its whole payload, so a very long page is slow for
the agent to produce in one go. Create the page with its first part, then add
the rest with envpilot_update_doc_draft (mode: "append"), or rewrite one
part with mode: "replace_section" and a section id. Every edit re-checks the
whole resulting body, so splitting text across calls cannot slip a secret
or an injected instruction past the scanner. Published pages cannot be edited
over MCP.
envpilot_get_doc returns at most 60,000 characters per call. Pass
view: "outline" for the section list, section for one section, and the
nextCursor from a response as cursor to keep reading. Breaks fall on
paragraph boundaries and never inside a code block, unless a single line or
code block is itself longer than 60,000 characters; that one is split at the
limit.
Retrying safely#
envpilot_create_doc accepts client_ref, a string the caller picks. A
second call with the same client_ref returns the first page instead of
creating a duplicate, as long as you can still reach that project. Use it when a call timed out and you cannot tell
whether it landed.
Pages name variables, never values#
A page may reference API_BASE_URL; it never contains the value. Resolving it
is the reader's own envpilot_get_variables call under their own role, so
reading documentation can never become reading secrets.
Content is scanned on write#
Every proposed page is checked before it is stored. Credential shapes (PEM blocks, provider key prefixes, connection strings with inline passwords) are rejected outright, and text instructing an agent to call a tool is refused. High-entropy strings that are normal in API documentation — sample JWTs, base64 payloads, commit SHAs — pass through untouched.
Cost of each call#
| Tool | Bucket | Audited |
|---|---|---|
list_projects, search, search_docs, get_doc, list_doc_drafts, list_files, list_accounts, list_requests, list_organizations | 120/min metadata | No |
get_variables, get_variable, get_account | 30/min value pulls | Yes |
get_file | File downloads, per person | Yes |
create_variable, update_variable, create_account | 30/min writes, per person | Yes |
upload_file | 20/min file uploads | Yes |
create_doc | 30/hour per person, burst 10 | Yes |
update_doc_draft | 120/hour per person, burst 40 | Yes |
Limits#
- A project you cannot see behaves exactly as if it does not exist.
envpilot_searchnever returns values, only names and keys.- There is no tool that deletes anything, publishes a documentation page, approves a request, or overrides a protected environment.
- Documentation search is full-text: whole words, with only the last term prefix-matched. It is not a substring match and does not correct typos.