MCP server

Revelica exposes your workspace over the Model Context Protocol, so your coding agent can read the product context your team has built and write what it learns back into the same place. Product context flows out, build context flows back.

Any MCP-capable client can connect. Claude, Claude Code, Cowork, Codex and Cursor are covered below, and the plugin ships for Claude Code and Codex.

Endpoint


Server URL	`https://api.revelica.com/mcp`
Transport	Streamable HTTP
Authentication	OAuth 2.0 with dynamic client registration
Prerequisite	A Revelica workspace (get Revelica free)

There is no API key to create or rotate. The first time a client connects it registers itself and opens a browser window for you to sign in. After that it holds a token scoped to your own account.

The server publishes its own metadata at /.well-known/mcp/server-card.json, and it is listed in the MCP registry as com.revelica/revelica.

Setup

Claude (web, desktop, and mobile)

Open Settings → Connectors → Add custom connector, paste the server URL, and authorize when the browser window opens.

https://api.revelica.com/mcp

Claude Code

Install the plugin. It registers the MCP server and adds Revelica's three starting skills in one step:

/plugin marketplace add revelica/skills
/plugin install revelica@revelica

To add the server on its own:

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

Run /mcp to sign in.

Cowork

Add the server URL as a custom connector in your organization's settings, the same way as Claude above.

Codex

Install the plugin from the same repository:

codex plugin marketplace add revelica/skills
codex plugin add revelica@revelica

Cursor

Add the server to .mcp.json in your project, or to your global Cursor MCP configuration:

{
  "mcpServers": {
    "revelica": {
      "type": "http",
      "url": "https://api.revelica.com/mcp"
    }
  }
}

The server's skills load with load_skill as in any client.

Other clients

Any client that speaks streamable HTTP and supports OAuth, including Gemini CLI, can connect with the server URL alone. Clients without OAuth support cannot connect, because Revelica does not issue static API keys.

Tools

Six tools, four of them read-only. Clients that understand tool annotations run the read-only ones without asking every time, and prompt before anything writes.

Tool	What it does	Access
`query`	Search or filter the workspace. Returns ranked matches, plus the schema template for a type when you filter by one.	Read
`read`	Open one record in full by id, or navigate into a subtree with a dotted field path. Read a story map and it comes back with every referenced story, assumption, segment, opportunity, and activity already joined in.	Read
`list_skills`	List the Revelica skills your workspace exposes to agents.	Read
`load_skill`	Load a skill's instructions, along with any workspace data it prefetches.	Read
`create`	Create a record, validated against the registered schema for its type.	Write
`update`	Update specific fields on an existing record using dot-path notation. Artifacts get a new version, entities update in place.	Write

What an agent can reach

Everything in your workspace, by type:

Customer research: interview snapshots with their sources cited, opportunities, customer segments, coverage
Market position: competitive analyses, market maps, company profiles, value propositions
Strategy: objectives, key results, metrics and measurements
Experience: user story maps, storyboards, tasks
Bets and evidence: ideas, hypotheses, assumptions, ACH matrices, experiments with their plans and reports, decisions, RICE assessments
Delivery: specs and documents, execution plans, projects

Filter by any of these with query(artifact_type=...). When you do, the response also carries the schema template for that type, so an agent knows the shape before it writes anything.

Skills

Skills are Revelica's playbooks: instructions an agent loads on demand with load_skill, along with a summary of the workspace data the playbook starts from. They live on the server, so every client gets the current version and the Revelica app runs the same ones.

The plugin adds three starting skills, the same starting points the app suggests. In Claude Code each is also a slash command:

Skill	What it does
`drive-product-outcome`	Plan the work toward your goal: read the goal, its measurements, the active bets and their evidence, then recommend the next action and save it as a task. Sets up anything the workspace is missing along the way.
`ingestion`	Save a URL, file, or note as evidence, with the findings pulled out.
`interview-snapshot`	Turn a customer interview transcript into a snapshot and the customer problems worth solving.

The server has more for the rest of the loop, including frame-bet, find-assumptions, design-experiment, competitive-analysis, write-product-spec, and product-for-coding-agents, which orients a coding agent in the product before it writes code and covers what to write back. The server's instructions tell the agent which skill fits which job, and list_skills shows everything your workspace exposes.

Every plugin manifest is public at github.com/revelica/skills.

Common use cases

Build from the spec, not from memory. "Read the idea and spec behind this ticket, then implement it. Flag anything in the spec you can't build as written." No more digging through chat history to remember what was decided.

Capture an idea mid-build. "Save this as an idea in Revelica with the problem and the technical decisions we just made." The idea lands in the workspace where your team and other agents can see it, instead of dying in the transcript.

Turn a customer conversation into structured research. "Here's the transcript from today's call. Create an interview snapshot, link it to the right customer segment, and list the opportunities it surfaces."

Check a claim against evidence. "Are we sure customers want this? Find every interview snapshot that mentions it and quote what they actually said."

Look at the market. "Summarize our competitive landscape for this feature area and tell me where our value proposition genuinely differs."

Find the risky assumptions. "Read the story map for this idea and list the assumptions we'd be betting on, worst first." The ones worth testing before anyone writes code.

Write the spec. "Use the write-product-spec skill to draft a spec for this idea."

Close the loop after shipping. "Write this up as an implementation doc on the idea, log the PR, and record the requirement we changed halfway through."

Permissions and security

Every request is signed in as you. Access is enforced in the database by row-level security, so an agent only ever sees and changes data in the workspace you belong to. Never another customer's, and never another workspace's inside your own company.

Revelica does not read your chat history, conversation summaries, or local files. Only what a tool is called with reaches the server.

Connect without a Revelica workspace and every tool returns an explanatory error pointing you at app.revelica.com, rather than failing silently.

See the Privacy Policy for data handling, retention, and sub-processors.

FAQ

Why won't it connect?

Clear the stored tokens for the server in your client and connect again. The client re-registers and gets fresh ones. In Claude Code, /mcp shows the connection state and lets you sign in again.

Do I need the plugin, or just the server?

The server alone gives you all six tools and every skill. The plugin registers the server for you and adds the three starting skills, so they show up as suggestions (and, in Claude Code, as slash commands).

I'm reviewing Revelica for a plugin directory

Test setup is in REVIEWING.md. The invite link for the demo workspace is in your review notes.